2013-01-18 04:45:11 +04:00
|
|
|
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
|
|
|
|
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
|
2012-05-21 15:12:37 +04:00
|
|
|
/* This Source Code Form is subject to the terms of the Mozilla Public
|
|
|
|
* License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
|
|
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
2010-08-12 23:37:44 +04:00
|
|
|
|
|
|
|
#include "prio.h"
|
2015-09-16 06:49:53 +03:00
|
|
|
#include "PLDHashTable.h"
|
2014-04-09 08:57:52 +04:00
|
|
|
#include "mozilla/IOInterposer.h"
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
#include "mozilla/AutoMemMap.h"
|
|
|
|
#include "mozilla/IOBuffers.h"
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
#include "mozilla/ipc/FileDescriptor.h"
|
|
|
|
#include "mozilla/ipc/GeckoChildProcessHost.h"
|
2013-06-23 16:03:39 +04:00
|
|
|
#include "mozilla/MemoryReporting.h"
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
#include "mozilla/MemUtils.h"
|
2020-04-11 00:16:15 +03:00
|
|
|
#include "mozilla/MmapFaultHandler.h"
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
#include "mozilla/ResultExtensions.h"
|
2010-08-12 23:37:44 +04:00
|
|
|
#include "mozilla/scache/StartupCache.h"
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
#include "mozilla/scache/StartupCacheChild.h"
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
#include "mozilla/ScopeExit.h"
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
#include "mozilla/dom/ContentChild.h"
|
|
|
|
#include "mozilla/dom/ContentParent.h"
|
|
|
|
#include "mozilla/dom/ipc/MemMapSnapshot.h"
|
2010-08-12 23:37:44 +04:00
|
|
|
|
|
|
|
#include "nsClassHashtable.h"
|
|
|
|
#include "nsComponentManagerUtils.h"
|
2019-03-05 19:52:57 +03:00
|
|
|
#include "nsCRT.h"
|
2010-08-12 23:37:44 +04:00
|
|
|
#include "nsDirectoryServiceUtils.h"
|
|
|
|
#include "nsIClassInfo.h"
|
|
|
|
#include "nsIFile.h"
|
|
|
|
#include "nsIObserver.h"
|
|
|
|
#include "nsIOutputStream.h"
|
|
|
|
#include "nsISupports.h"
|
|
|
|
#include "nsITimer.h"
|
2011-01-07 22:04:33 +03:00
|
|
|
#include "mozilla/Omnijar.h"
|
|
|
|
#include "prenv.h"
|
2012-02-22 23:03:52 +04:00
|
|
|
#include "mozilla/Telemetry.h"
|
2011-04-01 09:24:16 +04:00
|
|
|
#include "nsThreadUtils.h"
|
|
|
|
#include "nsXULAppAPI.h"
|
2011-07-01 01:58:03 +04:00
|
|
|
#include "nsIProtocolHandler.h"
|
2017-01-05 18:34:26 +03:00
|
|
|
#include "GeckoProfiler.h"
|
2020-06-24 19:00:32 +03:00
|
|
|
#include "nsAppRunner.h"
|
2011-04-01 09:24:16 +04:00
|
|
|
|
2020-04-11 00:16:15 +03:00
|
|
|
#if defined(XP_WIN)
|
|
|
|
# include <windows.h>
|
|
|
|
#endif
|
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
#ifdef IS_BIG_ENDIAN
|
|
|
|
# define SC_ENDIAN "big"
|
|
|
|
#else
|
|
|
|
# define SC_ENDIAN "little"
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#if PR_BYTES_PER_WORD == 4
|
|
|
|
# define SC_WORDSIZE "4"
|
|
|
|
#else
|
|
|
|
# define SC_WORDSIZE "8"
|
|
|
|
#endif
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
#define DOC_ELEM_INSERTED_TOPIC "document-element-inserted"
|
|
|
|
#define CONTENT_DOCUMENT_LOADED_TOPIC "content-document-loaded"
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
using namespace mozilla::Compression;
|
|
|
|
|
2011-12-19 04:20:36 +04:00
|
|
|
namespace mozilla {
|
|
|
|
namespace scache {
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// This is included here, rather than as a member of the StartupCache
|
|
|
|
// singleton, because including it there would mean we need to pull
|
|
|
|
// in the chromium IPC headers everywhere we reference StartupCache.h
|
|
|
|
static ipc::FileDescriptor sSharedDataFD;
|
|
|
|
|
2013-12-08 09:39:47 +04:00
|
|
|
MOZ_DEFINE_MALLOC_SIZE_OF(StartupCacheMallocSizeOf)
|
|
|
|
|
2013-11-07 09:35:30 +04:00
|
|
|
NS_IMETHODIMP
|
|
|
|
StartupCache::CollectReports(nsIHandleReportCallback* aHandleReport,
|
2014-05-21 10:06:54 +04:00
|
|
|
nsISupports* aData, bool aAnonymize) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
if (XRE_IsParentProcess()) {
|
|
|
|
MOZ_COLLECT_REPORT(
|
|
|
|
"explicit/startup-cache/mapping", KIND_NONHEAP, UNITS_BYTES,
|
|
|
|
mCacheData.nonHeapSizeOfExcludingThis(),
|
|
|
|
"Memory used to hold the mapping of the startup cache from file. "
|
|
|
|
"This memory is likely to be swapped out shortly after start-up.");
|
|
|
|
} else {
|
|
|
|
// In the child process, mCacheData actually points to the parent's
|
|
|
|
// mSharedData. We just want to report this once as explicit under
|
|
|
|
// the parent process's report.
|
|
|
|
MOZ_COLLECT_REPORT(
|
|
|
|
"startup-cache-shared-mapping", KIND_NONHEAP, UNITS_BYTES,
|
|
|
|
mCacheData.nonHeapSizeOfExcludingThis(),
|
|
|
|
"Memory used to hold the decompressed contents of part of the "
|
|
|
|
"startup cache, to be used by child processes. This memory is "
|
|
|
|
"likely to be swapped out shortly after start-up.");
|
|
|
|
}
|
2016-08-24 08:23:45 +03:00
|
|
|
|
|
|
|
MOZ_COLLECT_REPORT("explicit/startup-cache/data", KIND_HEAP, UNITS_BYTES,
|
|
|
|
HeapSizeOfIncludingThis(StartupCacheMallocSizeOf),
|
|
|
|
"Memory used by the startup cache for things other than "
|
|
|
|
"the file mapping.");
|
2011-12-19 04:20:36 +04:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MOZ_COLLECT_REPORT(
|
|
|
|
"explicit/startup-cache/shared-mapping", KIND_NONHEAP, UNITS_BYTES,
|
|
|
|
mSharedData.nonHeapSizeOfExcludingThis(),
|
|
|
|
"Memory used to hold the decompressed contents of part of the "
|
|
|
|
"startup cache, to be used by child processes. This memory is "
|
|
|
|
"likely to be swapped out shortly after start-up.");
|
|
|
|
|
2013-11-07 09:35:30 +04:00
|
|
|
return NS_OK;
|
2014-04-15 11:28:21 +04:00
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
static const uint8_t MAGIC[] = "startupcache0002";
|
|
|
|
// This is a heuristic value for how much to reserve for mTable to avoid
|
|
|
|
// rehashing. This is not a hard limit in release builds, but it is in
|
|
|
|
// debug builds as it should be stable. If we exceed this number we should
|
|
|
|
// just increase it.
|
2019-10-04 23:45:18 +03:00
|
|
|
static const size_t STARTUP_CACHE_RESERVE_CAPACITY = 450;
|
|
|
|
// This is a hard limit which we will assert on, to ensure that we don't
|
|
|
|
// have some bug causing runaway cache growth.
|
|
|
|
static const size_t STARTUP_CACHE_MAX_CAPACITY = 5000;
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
|
2020-06-24 19:00:32 +03:00
|
|
|
// Not const because we change it for gtests.
|
|
|
|
static uint8_t STARTUP_CACHE_WRITE_TIMEOUT = 60;
|
|
|
|
|
2015-05-16 11:07:10 +03:00
|
|
|
#define STARTUP_CACHE_NAME "startupCache." SC_WORDSIZE "." SC_ENDIAN
|
2013-11-02 03:21:33 +04:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
static inline Result<Ok, nsresult> Write(PRFileDesc* fd, const void* data,
|
|
|
|
int32_t len) {
|
|
|
|
if (PR_Write(fd, data, len) != len) {
|
|
|
|
return Err(NS_ERROR_FAILURE);
|
|
|
|
}
|
|
|
|
return Ok();
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline Result<Ok, nsresult> Seek(PRFileDesc* fd, int32_t offset) {
|
|
|
|
if (PR_Seek(fd, offset, PR_SEEK_SET) == -1) {
|
|
|
|
return Err(NS_ERROR_FAILURE);
|
|
|
|
}
|
|
|
|
return Ok();
|
|
|
|
}
|
|
|
|
|
|
|
|
static nsresult MapLZ4ErrorToNsresult(size_t aError) {
|
|
|
|
return NS_ERROR_FAILURE;
|
|
|
|
}
|
|
|
|
|
2020-07-01 22:16:28 +03:00
|
|
|
StartupCache* StartupCache::GetSingleton() {
|
|
|
|
return StartupCache::gStartupCache;
|
2020-07-01 20:07:29 +03:00
|
|
|
}
|
|
|
|
|
2020-07-01 22:16:28 +03:00
|
|
|
void StartupCache::DeleteSingleton() { StartupCache::gStartupCache = nullptr; }
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
|
2020-07-02 05:49:49 +03:00
|
|
|
nsresult StartupCache::PartialInitSingleton(nsIFile* aProfileLocalDir) {
|
|
|
|
#ifdef MOZ_DISABLE_STARTUPCACHE
|
|
|
|
return NS_OK;
|
|
|
|
#else
|
|
|
|
if (!XRE_IsParentProcess()) {
|
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
nsresult rv;
|
|
|
|
StartupCache::gStartupCache = new StartupCache();
|
|
|
|
|
2020-07-02 05:49:49 +03:00
|
|
|
rv = StartupCache::gStartupCache->PartialInit(aProfileLocalDir);
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
if (NS_FAILED(rv)) {
|
|
|
|
StartupCache::gStartupCache = nullptr;
|
|
|
|
}
|
|
|
|
return rv;
|
2020-07-02 05:49:49 +03:00
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
nsresult StartupCache::InitChildSingleton(char* aScacheHandleStr,
|
|
|
|
char* aScacheSizeStr) {
|
|
|
|
#ifdef MOZ_DISABLE_STARTUPCACHE
|
|
|
|
return NS_OK;
|
|
|
|
#else
|
|
|
|
MOZ_ASSERT(!XRE_IsParentProcess());
|
|
|
|
|
|
|
|
nsresult rv;
|
|
|
|
StartupCache::gStartupCache = new StartupCache();
|
|
|
|
|
|
|
|
rv = StartupCache::gStartupCache->ParseStartupCacheCmdLineArgs(
|
|
|
|
aScacheHandleStr, aScacheSizeStr);
|
|
|
|
if (NS_FAILED(rv)) {
|
|
|
|
StartupCache::gStartupCache = nullptr;
|
|
|
|
}
|
|
|
|
return rv;
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2020-07-02 05:49:49 +03:00
|
|
|
nsresult StartupCache::FullyInitSingleton() {
|
|
|
|
if (!StartupCache::gStartupCache) {
|
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
return StartupCache::gStartupCache->FullyInit();
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
2013-11-26 03:57:40 +04:00
|
|
|
StaticRefPtr<StartupCache> StartupCache::gStartupCache;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
ProcessType sProcessType;
|
2011-09-29 10:19:26 +04:00
|
|
|
bool StartupCache::gShutdownInitiated;
|
2012-10-11 12:17:15 +04:00
|
|
|
bool StartupCache::gIgnoreDiskCache;
|
2020-04-02 22:40:39 +03:00
|
|
|
bool StartupCache::gFoundDiskCacheOnInit;
|
2010-08-12 23:37:44 +04:00
|
|
|
|
2014-04-27 11:06:00 +04:00
|
|
|
NS_IMPL_ISUPPORTS(StartupCache, nsIMemoryReporter)
|
2013-11-26 03:57:40 +04:00
|
|
|
|
2013-01-18 04:45:11 +04:00
|
|
|
StartupCache::StartupCache()
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
: mLock("StartupCache::mLock"),
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
mDirty(false),
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
mWrittenOnce(false),
|
|
|
|
mCurTableReferenced(false),
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mLoaded(false),
|
|
|
|
mFullyInitialized(false),
|
2019-10-04 23:45:18 +03:00
|
|
|
mRequestedCount(0),
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mPrefetchSize(0),
|
|
|
|
mSharedDataSize(0),
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
mCacheEntriesBaseOffset(0),
|
2019-10-04 23:45:32 +03:00
|
|
|
mPrefetchThread(nullptr) {}
|
2010-08-12 23:37:44 +04:00
|
|
|
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
StartupCache::~StartupCache() { UnregisterWeakMemoryReporter(this); }
|
2010-08-12 23:37:44 +04:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
nsresult StartupCache::InitChild(StartupCacheChild* cacheChild) {
|
|
|
|
mChildActor = cacheChild;
|
|
|
|
sProcessType =
|
|
|
|
GetChildProcessType(dom::ContentChild::GetSingleton()->GetRemoteType());
|
|
|
|
|
|
|
|
mObserverService = do_GetService("@mozilla.org/observer-service;1");
|
|
|
|
|
|
|
|
if (!mObserverService) {
|
|
|
|
NS_WARNING("Could not get observerService.");
|
|
|
|
return NS_ERROR_UNEXPECTED;
|
|
|
|
}
|
|
|
|
|
|
|
|
mListener = new StartupCacheListener();
|
|
|
|
nsresult rv = mObserverService->AddObserver(
|
|
|
|
mListener, NS_XPCOM_SHUTDOWN_OBSERVER_ID, false);
|
|
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
|
|
|
rv = mObserverService->AddObserver(mListener, "startupcache-invalidate",
|
|
|
|
false);
|
|
|
|
|
|
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
|
|
|
|
|
|
|
if (mChildActor) {
|
|
|
|
if (sProcessType == ProcessType::PrivilegedAbout) {
|
|
|
|
// Since we control all of the documents loaded in the privileged
|
|
|
|
// content process, we can increase the window of active time for the
|
|
|
|
// StartupCache to include the scripts that are loaded until the
|
|
|
|
// first document finishes loading.
|
|
|
|
mContentStartupFinishedTopic.AssignLiteral(CONTENT_DOCUMENT_LOADED_TOPIC);
|
|
|
|
} else {
|
|
|
|
// In the child process, we need to freeze the startup cache before any
|
|
|
|
// untrusted code has been executed. The insertion of the first DOM
|
|
|
|
// document element may sometimes be earlier than is ideal, but at
|
|
|
|
// least it should always be safe.
|
|
|
|
mContentStartupFinishedTopic.AssignLiteral(DOC_ELEM_INSERTED_TOPIC);
|
|
|
|
}
|
|
|
|
|
|
|
|
mObserverService->AddObserver(mListener, mContentStartupFinishedTopic.get(),
|
|
|
|
false);
|
|
|
|
}
|
|
|
|
|
|
|
|
mFullyInitialized = true;
|
|
|
|
RegisterWeakMemoryReporter(this);
|
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
2020-07-02 05:49:49 +03:00
|
|
|
nsresult StartupCache::PartialInit(nsIFile* aProfileLocalDir) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(mLock);
|
2011-07-01 01:58:03 +04:00
|
|
|
// workaround for bug 653936
|
|
|
|
nsCOMPtr<nsIProtocolHandler> jarInitializer(
|
|
|
|
do_GetService(NS_NETWORK_PROTOCOL_CONTRACTID_PREFIX "jar"));
|
2013-01-18 04:45:11 +04:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
sProcessType = ProcessType::Parent;
|
2010-08-12 23:37:44 +04:00
|
|
|
nsresult rv;
|
|
|
|
|
2020-06-24 19:00:32 +03:00
|
|
|
if (mozilla::RunningGTest()) {
|
|
|
|
STARTUP_CACHE_WRITE_TIMEOUT = 3;
|
|
|
|
}
|
|
|
|
|
2011-01-07 22:04:33 +03:00
|
|
|
// This allows to override the startup cache filename
|
|
|
|
// which is useful from xpcshell, when there is no ProfLDS directory to keep
|
|
|
|
// cache in.
|
|
|
|
char* env = PR_GetEnv("MOZ_STARTUP_CACHE");
|
2016-11-11 19:57:08 +03:00
|
|
|
if (env && *env) {
|
2011-10-17 18:59:28 +04:00
|
|
|
rv = NS_NewLocalFile(NS_ConvertUTF8toUTF16(env), false,
|
|
|
|
getter_AddRefs(mFile));
|
2020-07-02 05:49:49 +03:00
|
|
|
} else if (aProfileLocalDir) {
|
2011-01-07 22:04:33 +03:00
|
|
|
nsCOMPtr<nsIFile> file;
|
2020-07-02 05:49:49 +03:00
|
|
|
rv = aProfileLocalDir->Clone(getter_AddRefs(file));
|
2011-01-07 22:04:33 +03:00
|
|
|
if (NS_FAILED(rv)) {
|
|
|
|
// return silently, this will fail in mochitests's xpcshell process.
|
|
|
|
return rv;
|
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
|
2020-07-01 11:29:29 +03:00
|
|
|
rv = file->AppendNative("startupCache"_ns);
|
2011-01-07 22:04:33 +03:00
|
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
2010-08-12 23:37:44 +04:00
|
|
|
|
2011-01-07 22:04:33 +03:00
|
|
|
// Try to create the directory if it's not there yet
|
|
|
|
rv = file->Create(nsIFile::DIRECTORY_TYPE, 0777);
|
|
|
|
if (NS_FAILED(rv) && rv != NS_ERROR_FILE_ALREADY_EXISTS) return rv;
|
|
|
|
|
2020-07-01 11:29:29 +03:00
|
|
|
rv = file->AppendNative(nsLiteralCString(STARTUP_CACHE_NAME));
|
2011-01-07 22:04:33 +03:00
|
|
|
|
|
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
2013-01-18 04:45:11 +04:00
|
|
|
|
2018-10-02 00:38:01 +03:00
|
|
|
mFile = file;
|
2020-07-02 05:49:49 +03:00
|
|
|
} else {
|
|
|
|
return NS_ERROR_INVALID_ARG;
|
2011-01-07 22:04:33 +03:00
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
|
|
|
|
NS_ENSURE_TRUE(mFile, NS_ERROR_UNEXPECTED);
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mDecompressionContext = MakeUnique<LZ4FrameDecompressionContext>(true);
|
|
|
|
|
|
|
|
Unused << mCacheData.init(mFile);
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
auto result = LoadArchive();
|
|
|
|
rv = result.isErr() ? result.unwrapErr() : NS_OK;
|
|
|
|
|
2020-07-01 22:16:28 +03:00
|
|
|
gFoundDiskCacheOnInit = rv != NS_ERROR_FILE_NOT_FOUND;
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
// Sometimes we don't have a cache yet, that's ok.
|
|
|
|
// If it's corrupted, just remove it and start over.
|
|
|
|
if (gIgnoreDiskCache || (NS_FAILED(rv) && rv != NS_ERROR_FILE_NOT_FOUND)) {
|
|
|
|
NS_WARNING("Failed to load startupcache file correctly, removing!");
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
InvalidateCacheImpl();
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
}
|
|
|
|
|
2020-07-01 22:16:28 +03:00
|
|
|
return NS_OK;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
}
|
|
|
|
|
2020-07-02 05:49:49 +03:00
|
|
|
nsresult StartupCache::FullyInit() {
|
|
|
|
mObserverService = do_GetService("@mozilla.org/observer-service;1");
|
|
|
|
|
|
|
|
if (!mObserverService) {
|
|
|
|
NS_WARNING("Could not get observerService.");
|
|
|
|
return NS_ERROR_UNEXPECTED;
|
|
|
|
}
|
|
|
|
|
|
|
|
mListener = new StartupCacheListener();
|
|
|
|
nsresult rv = mObserverService->AddObserver(
|
|
|
|
mListener, NS_XPCOM_SHUTDOWN_OBSERVER_ID, false);
|
|
|
|
if (NS_WARN_IF(NS_FAILED(rv))) {
|
|
|
|
return rv;
|
|
|
|
}
|
|
|
|
rv = mObserverService->AddObserver(mListener, "startupcache-invalidate",
|
|
|
|
false);
|
|
|
|
if (NS_WARN_IF(NS_FAILED(rv))) {
|
|
|
|
return rv;
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
RegisterWeakMemoryReporter(this);
|
|
|
|
mFullyInitialized = true;
|
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
void StartupCache::InitContentChild(dom::ContentParent& parent) {
|
|
|
|
auto* cache = GetSingleton();
|
|
|
|
if (!cache) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
auto processType = GetChildProcessType(parent.GetRemoteType());
|
|
|
|
bool wantScriptData = !cache->mInitializedProcesses.contains(processType);
|
|
|
|
cache->mInitializedProcesses += processType;
|
|
|
|
Unused << parent.SendPStartupCacheConstructor(wantScriptData);
|
|
|
|
}
|
|
|
|
|
|
|
|
void StartupCache::AddStartupCacheCmdLineArgs(
|
|
|
|
mozilla::ipc::GeckoChildProcessHost& procHost,
|
|
|
|
std::vector<std::string>& aExtraOpts) {
|
|
|
|
#ifndef ANDROID
|
|
|
|
// Don't send original cache data to new processes if the cache has been
|
|
|
|
// invalidated.
|
|
|
|
if (!mSharedData.initialized() || gIgnoreDiskCache) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Formats a pointer or pointer-sized-integer as a string suitable for
|
|
|
|
// passing in an arguments list.
|
|
|
|
auto formatPtrArg = [](auto arg) {
|
|
|
|
return nsPrintfCString("%zu", uintptr_t(arg));
|
|
|
|
};
|
|
|
|
if (!mSharedDataHandle) {
|
|
|
|
auto fd = mSharedData.cloneHandle();
|
|
|
|
MOZ_ASSERT(fd.IsValid());
|
|
|
|
|
|
|
|
mSharedDataHandle = fd.TakePlatformHandle();
|
|
|
|
}
|
|
|
|
|
|
|
|
# if defined(XP_WIN)
|
|
|
|
procHost.AddHandleToShare(mSharedDataHandle.get());
|
|
|
|
aExtraOpts.push_back("-scacheHandle");
|
|
|
|
aExtraOpts.push_back(formatPtrArg(mSharedDataHandle.get()).get());
|
|
|
|
# else
|
|
|
|
procHost.AddFdToRemap(mSharedDataHandle.get(), kStartupCacheFd);
|
|
|
|
# endif
|
|
|
|
|
|
|
|
aExtraOpts.push_back("-scacheSize");
|
|
|
|
aExtraOpts.push_back(formatPtrArg(mSharedDataSize).get());
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
nsresult StartupCache::ParseStartupCacheCmdLineArgs(char* aScacheHandleStr,
|
|
|
|
char* aScacheSizeStr) {
|
|
|
|
// Parses an arg containing a pointer-sized-integer.
|
|
|
|
auto parseUIntPtrArg = [](char*& aArg) {
|
|
|
|
// ContentParent uses %zu to print a word-sized unsigned integer. So
|
|
|
|
// even though strtoull() returns a long long int, it will fit in a
|
|
|
|
// uintptr_t.
|
|
|
|
return uintptr_t(strtoull(aArg, &aArg, 10));
|
|
|
|
};
|
|
|
|
|
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
if (aScacheHandleStr) {
|
|
|
|
#ifdef XP_WIN
|
|
|
|
auto parseHandleArg = [&](char*& aArg) {
|
|
|
|
return HANDLE(parseUIntPtrArg(aArg));
|
|
|
|
};
|
|
|
|
|
|
|
|
sSharedDataFD = FileDescriptor(parseHandleArg(aScacheHandleStr));
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef XP_UNIX
|
|
|
|
sSharedDataFD = FileDescriptor(UniqueFileHandle(kStartupCacheFd));
|
|
|
|
#endif
|
|
|
|
MOZ_TRY(mCacheData.initWithHandle(sSharedDataFD,
|
|
|
|
parseUIntPtrArg(aScacheSizeStr)));
|
|
|
|
}
|
|
|
|
|
|
|
|
nsresult rv;
|
|
|
|
auto result = LoadArchive();
|
|
|
|
rv = result.isErr() ? result.unwrapErr() : NS_OK;
|
|
|
|
|
|
|
|
// Sometimes we don't have a cache yet, that's ok.
|
|
|
|
// If it's corrupted, just remove it and start over.
|
|
|
|
if (gIgnoreDiskCache || (NS_FAILED(rv) && rv != NS_ERROR_FILE_NOT_FOUND)) {
|
|
|
|
NS_WARNING("Failed to load startupcache file correctly, removing!");
|
|
|
|
InvalidateCacheImpl();
|
|
|
|
}
|
|
|
|
|
2020-07-02 05:49:49 +03:00
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
ProcessType StartupCache::GetChildProcessType(const nsAString& remoteType) {
|
|
|
|
if (remoteType.EqualsLiteral(EXTENSION_REMOTE_TYPE)) {
|
|
|
|
return ProcessType::Extension;
|
|
|
|
}
|
|
|
|
if (remoteType.EqualsLiteral(PRIVILEGEDABOUT_REMOTE_TYPE)) {
|
|
|
|
return ProcessType::PrivilegedAbout;
|
|
|
|
}
|
|
|
|
return ProcessType::Web;
|
|
|
|
}
|
|
|
|
|
2019-10-04 23:45:32 +03:00
|
|
|
void StartupCache::StartPrefetchMemoryThread() {
|
|
|
|
// XXX: It would be great for this to not create its own thread, unfortunately
|
|
|
|
// there doesn't seem to be an existing thread that makes sense for this, so
|
|
|
|
// barring a coordinated global scheduling system this is the best we get.
|
|
|
|
mPrefetchThread = PR_CreateThread(
|
|
|
|
PR_USER_THREAD, StartupCache::ThreadedPrefetch, this, PR_PRIORITY_NORMAL,
|
2019-10-04 23:45:41 +03:00
|
|
|
PR_GLOBAL_THREAD, PR_JOINABLE_THREAD, 256 * 1024);
|
2019-10-04 23:45:32 +03:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
Result<Ok, nsresult> StartupCache::LoadEntriesOffDisk() {
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
auto size = mCacheData.size();
|
2019-05-21 20:32:45 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
uint32_t headerSize;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
uint32_t prefetchSize;
|
|
|
|
size_t headerOffset =
|
|
|
|
sizeof(MAGIC) + sizeof(headerSize) + sizeof(prefetchSize);
|
|
|
|
if (size < headerOffset) {
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-09-28 01:15:35 +03:00
|
|
|
}
|
2011-12-08 14:03:36 +04:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
auto data = mCacheData.get<uint8_t>();
|
|
|
|
auto end = data + size;
|
2011-12-08 14:03:36 +04:00
|
|
|
|
2020-04-11 00:16:15 +03:00
|
|
|
MMAP_FAULT_HANDLER_BEGIN_BUFFER(data.get(), size)
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
if (memcmp(MAGIC, data.get(), sizeof(MAGIC))) {
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
data += sizeof(MAGIC);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-09-28 01:15:35 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
headerSize = LittleEndian::readUint32(data.get());
|
|
|
|
data += sizeof(headerSize);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-09-28 01:15:35 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
prefetchSize = LittleEndian::readUint32(data.get());
|
|
|
|
data += sizeof(prefetchSize);
|
|
|
|
|
|
|
|
if (size < prefetchSize) {
|
2019-10-22 06:15:18 +03:00
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-09-28 01:15:35 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (size < headerOffset + headerSize) {
|
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (CanPrefetchMemory()) {
|
|
|
|
mPrefetchSize = prefetchSize;
|
|
|
|
StartPrefetchMemoryThread();
|
|
|
|
}
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
Range<uint8_t> header(data, data + headerSize);
|
|
|
|
data += headerSize;
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
uint32_t numSharedEntries = 0;
|
|
|
|
|
|
|
|
mCacheEntriesBaseOffset = headerOffset + headerSize;
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
{
|
2019-10-04 23:45:18 +03:00
|
|
|
if (!mTable.reserve(STARTUP_CACHE_RESERVE_CAPACITY)) {
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
auto cleanup = MakeScopeExit([&]() {
|
2020-04-11 00:16:07 +03:00
|
|
|
WaitOnPrefetchThread();
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
mTable.clear();
|
|
|
|
mCacheData.reset();
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mSharedDataSize = 0;
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
});
|
|
|
|
loader::InputBuffer buf(header);
|
|
|
|
|
|
|
|
uint32_t currentOffset = 0;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mSharedDataSize = sizeof(MAGIC) + sizeof(numSharedEntries);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
while (!buf.finished()) {
|
|
|
|
uint32_t offset = 0;
|
|
|
|
uint32_t compressedSize = 0;
|
|
|
|
uint32_t uncompressedSize = 0;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
uint8_t sharedToChild = 0;
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
nsCString key;
|
|
|
|
buf.codeUint32(offset);
|
|
|
|
buf.codeUint32(compressedSize);
|
|
|
|
buf.codeUint32(uncompressedSize);
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
buf.codeUint8(sharedToChild);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
buf.codeString(key);
|
|
|
|
|
2019-10-22 06:15:18 +03:00
|
|
|
if (offset + compressedSize > end - data) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (XRE_IsParentProcess()) {
|
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
// If we're not the parent process, then we received a buffer with
|
|
|
|
// only data that was already prefetched. If that's the case, then
|
|
|
|
// just break because we've hit the end of that range.
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
EnumSet<StartupCacheEntryFlags> entryFlags;
|
|
|
|
if (sharedToChild) {
|
|
|
|
// We encode the key as a u16 indicating the length, followed by the
|
|
|
|
// contents of the string. We encode the uncompressed data similarly,
|
|
|
|
// but with a u32 indicating the length rather than a u16.
|
|
|
|
size_t sharedEntrySize = sizeof(uint16_t) + key.Length() +
|
|
|
|
sizeof(uint32_t) + uncompressedSize;
|
|
|
|
numSharedEntries++;
|
|
|
|
mSharedDataSize += sharedEntrySize;
|
|
|
|
entryFlags += StartupCacheEntryFlags::Shared;
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// Make sure offsets match what we'd expect based on script ordering and
|
|
|
|
// size, as a basic sanity check.
|
|
|
|
if (offset != currentOffset) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
currentOffset += compressedSize;
|
|
|
|
|
2019-12-30 21:50:56 +03:00
|
|
|
// We could use mTable.putNew if we knew the file we're loading weren't
|
|
|
|
// corrupt. However, we don't know that, so check if the key already
|
|
|
|
// exists. If it does, we know the file must be corrupt.
|
|
|
|
decltype(mTable)::AddPtr p = mTable.lookupForAdd(key);
|
|
|
|
if (p) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
2019-12-30 21:50:56 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!mTable.add(p, key,
|
|
|
|
StartupCacheEntry(offset, compressedSize,
|
|
|
|
uncompressedSize, entryFlags))) {
|
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (buf.error()) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MOZ_ASSERT(false, "StartupCache file is corrupt.");
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
2011-12-08 14:03:36 +04:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
cleanup.release();
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// We only get to init mSharedData once - we can't edit it after we created
|
|
|
|
// it, otherwise we'd be pulling memory out from underneath our child
|
|
|
|
// processes.
|
|
|
|
if (!mSharedData.initialized()) {
|
|
|
|
ipc::MemMapSnapshot mem;
|
|
|
|
if (mem.Init(mSharedDataSize).isErr()) {
|
|
|
|
MOZ_ASSERT(false, "Failed to init shared StartupCache data.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
auto ptr = mem.Get<uint8_t>();
|
|
|
|
Range<uint8_t> ptrRange(ptr, ptr + mSharedDataSize);
|
|
|
|
loader::PreallocatedOutputBuffer sharedBuf(ptrRange);
|
|
|
|
|
|
|
|
if (sizeof(MAGIC) > sharedBuf.remainingCapacity()) {
|
|
|
|
MOZ_ASSERT(false);
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
uint8_t* sharedMagic = sharedBuf.write(sizeof(MAGIC));
|
|
|
|
memcpy(sharedMagic, MAGIC, sizeof(MAGIC));
|
|
|
|
sharedBuf.codeUint32(numSharedEntries);
|
|
|
|
|
|
|
|
for (auto iter = mTable.iter(); !iter.done(); iter.next()) {
|
|
|
|
const auto& key = iter.get().key();
|
|
|
|
auto& value = iter.get().value();
|
|
|
|
|
|
|
|
if (!value.mFlags.contains(StartupCacheEntryFlags::Shared)) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
sharedBuf.codeString(key);
|
|
|
|
sharedBuf.codeUint32(value.mUncompressedSize);
|
|
|
|
|
|
|
|
if (value.mUncompressedSize > sharedBuf.remainingCapacity()) {
|
|
|
|
MOZ_ASSERT(false);
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
value.mData =
|
|
|
|
MaybeOwnedCharPtr((char*)(sharedBuf.write(value.mUncompressedSize)));
|
|
|
|
|
|
|
|
if (sharedBuf.error()) {
|
|
|
|
MOZ_ASSERT(false, "Failed serializing to shared memory.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
MOZ_TRY(DecompressEntry(value));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mem.Finalize(mSharedData).isErr()) {
|
|
|
|
MOZ_ASSERT(false, "Failed to freeze shared StartupCache data.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-04-11 00:16:15 +03:00
|
|
|
MMAP_FAULT_HANDLER_CATCH(Err(NS_ERROR_UNEXPECTED))
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Ok();
|
|
|
|
}
|
2019-09-28 01:15:01 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
Result<Ok, nsresult> StartupCache::LoadEntriesFromSharedMemory() {
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
|
|
|
auto size = mCacheData.size();
|
|
|
|
auto start = mCacheData.get<uint8_t>();
|
|
|
|
auto end = start + size;
|
|
|
|
|
|
|
|
Range<uint8_t> range(start, end);
|
|
|
|
loader::InputBuffer buf(range);
|
|
|
|
|
|
|
|
if (sizeof(MAGIC) > buf.remainingCapacity()) {
|
|
|
|
MOZ_ASSERT(false, "Bad shared StartupCache buffer.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
const uint8_t* magic = buf.read(sizeof(MAGIC));
|
|
|
|
if (memcmp(magic, MAGIC, sizeof(MAGIC)) != 0) {
|
|
|
|
MOZ_ASSERT(false, "Bad shared StartupCache buffer.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
uint32_t numEntries = 0;
|
|
|
|
if (!buf.codeUint32(numEntries) || numEntries > STARTUP_CACHE_MAX_CAPACITY) {
|
|
|
|
MOZ_ASSERT(false, "Bad number of entries in shared StartupCache buffer.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!mTable.reserve(STARTUP_CACHE_RESERVE_CAPACITY)) {
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
uint32_t entriesSeen = 0;
|
|
|
|
while (!buf.finished()) {
|
|
|
|
entriesSeen++;
|
|
|
|
if (entriesSeen > numEntries) {
|
|
|
|
MOZ_ASSERT(false, "More entries than expected in StartupCache buffer.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
nsCString key;
|
|
|
|
uint32_t uncompressedSize = 0;
|
|
|
|
buf.codeString(key);
|
|
|
|
buf.codeUint32(uncompressedSize);
|
|
|
|
|
|
|
|
if (uncompressedSize > buf.remainingCapacity()) {
|
|
|
|
MOZ_ASSERT(false,
|
|
|
|
"StartupCache entry larger than remaining buffer size.");
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
const char* data =
|
|
|
|
reinterpret_cast<const char*>(buf.read(uncompressedSize));
|
|
|
|
|
|
|
|
StartupCacheEntry entry(0, 0, uncompressedSize,
|
|
|
|
StartupCacheEntryFlags::Shared);
|
|
|
|
// The const cast is unfortunate here. We have runtime guarantees via memory
|
|
|
|
// protections that mData will not be modified, but we do not have compile
|
|
|
|
// time guarantees.
|
|
|
|
entry.mData = MaybeOwnedCharPtr(const_cast<char*>(data));
|
|
|
|
|
|
|
|
if (!mTable.putNew(key, std::move(entry))) {
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return Ok();
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* LoadArchive can only be called from the main thread.
|
|
|
|
*/
|
|
|
|
Result<Ok, nsresult> StartupCache::LoadArchive() {
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
|
|
|
|
|
|
|
if (gIgnoreDiskCache) {
|
|
|
|
return Err(NS_ERROR_FAILURE);
|
|
|
|
}
|
|
|
|
|
|
|
|
mLoaded = true;
|
|
|
|
|
|
|
|
if (!mCacheData.initialized()) {
|
|
|
|
return Err(NS_ERROR_FILE_NOT_FOUND);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (XRE_IsParentProcess()) {
|
|
|
|
MOZ_TRY(LoadEntriesOffDisk());
|
|
|
|
} else {
|
|
|
|
MOZ_TRY(LoadEntriesFromSharedMemory());
|
|
|
|
}
|
|
|
|
|
|
|
|
return Ok();
|
|
|
|
}
|
|
|
|
|
|
|
|
Result<Ok, nsresult> StartupCache::DecompressEntry(StartupCacheEntry& aEntry) {
|
|
|
|
MOZ_ASSERT(XRE_IsParentProcess());
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
|
|
|
|
|
|
|
size_t totalRead = 0;
|
|
|
|
size_t totalWritten = 0;
|
|
|
|
Span<const char> compressed = MakeSpan(
|
|
|
|
mCacheData.get<char>().get() + mCacheEntriesBaseOffset + aEntry.mOffset,
|
|
|
|
aEntry.mCompressedSize);
|
|
|
|
Span<char> uncompressed =
|
|
|
|
MakeSpan(aEntry.mData.get(), aEntry.mUncompressedSize);
|
|
|
|
bool finished = false;
|
|
|
|
while (!finished) {
|
|
|
|
auto result = mDecompressionContext->Decompress(
|
|
|
|
uncompressed.From(totalWritten), compressed.From(totalRead));
|
|
|
|
if (NS_WARN_IF(result.isErr())) {
|
|
|
|
aEntry.mData = nullptr;
|
|
|
|
InvalidateCacheImpl();
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
auto decompressionResult = result.unwrap();
|
|
|
|
totalRead += decompressionResult.mSizeRead;
|
|
|
|
totalWritten += decompressionResult.mSizeWritten;
|
|
|
|
finished = decompressionResult.mFinished;
|
|
|
|
}
|
|
|
|
|
|
|
|
return Ok();
|
|
|
|
}
|
|
|
|
|
2019-10-04 23:44:32 +03:00
|
|
|
bool StartupCache::HasEntry(const char* id) {
|
|
|
|
AUTO_PROFILER_LABEL("StartupCache::HasEntry", OTHER);
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(mLock);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return mTable.has(nsDependentCString(id));
|
2019-10-04 23:44:32 +03:00
|
|
|
}
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
nsresult StartupCache::GetBuffer(const char* id, const char** outbuf,
|
2016-02-18 20:26:28 +03:00
|
|
|
uint32_t* length) {
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
Telemetry::LABELS_STARTUP_CACHE_REQUESTS label =
|
|
|
|
Telemetry::LABELS_STARTUP_CACHE_REQUESTS::Miss;
|
|
|
|
auto telemetry =
|
|
|
|
MakeScopeExit([&label] { Telemetry::AccumulateCategorical(label); });
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
if (!mLoaded) {
|
|
|
|
return NS_ERROR_NOT_AVAILABLE;
|
|
|
|
}
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
decltype(mTable)::Ptr p = mTable.lookup(nsDependentCString(id));
|
|
|
|
if (!p) {
|
|
|
|
return NS_ERROR_NOT_AVAILABLE;
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// Track that something holds a reference into mTable, so we know to hold
|
|
|
|
// onto it in case the cache is invalidated.
|
|
|
|
mCurTableReferenced = true;
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
auto& value = p->value();
|
|
|
|
if (value.mData) {
|
|
|
|
label = Telemetry::LABELS_STARTUP_CACHE_REQUESTS::HitMemory;
|
|
|
|
} else {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// We do not decompress entries in child processes: we receive them
|
|
|
|
// uncompressed in shared memory.
|
|
|
|
if (!XRE_IsParentProcess() || !mCacheData.initialized()) {
|
2020-07-01 22:16:28 +03:00
|
|
|
return NS_ERROR_NOT_AVAILABLE;
|
|
|
|
}
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
value.mData = MaybeOwnedCharPtr(value.mUncompressedSize);
|
|
|
|
MMAP_FAULT_HANDLER_BEGIN_BUFFER(value.mData.get(), value.mUncompressedSize)
|
|
|
|
|
|
|
|
auto decompressionResult = DecompressEntry(value);
|
|
|
|
if (decompressionResult.isErr()) {
|
|
|
|
return decompressionResult.unwrapErr();
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-09-28 01:15:35 +03:00
|
|
|
}
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
|
2020-04-11 00:16:15 +03:00
|
|
|
MMAP_FAULT_HANDLER_CATCH(NS_ERROR_FAILURE)
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
label = Telemetry::LABELS_STARTUP_CACHE_REQUESTS::HitDisk;
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (value.mRequestedOrder == kStartupCacheEntryNotRequested) {
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
value.mRequestedOrder = ++mRequestedCount;
|
2019-10-04 23:45:18 +03:00
|
|
|
MOZ_ASSERT(mRequestedCount <= mTable.count(),
|
|
|
|
"Somehow we requested more StartupCache items than exist.");
|
|
|
|
ResetStartupWriteTimerCheckingReadCount();
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
}
|
|
|
|
|
2019-10-04 23:45:18 +03:00
|
|
|
*outbuf = value.mData.get();
|
|
|
|
*length = value.mUncompressedSize;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return NS_OK;
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// Client gives ownership of inbuf.
|
2018-02-17 02:30:47 +03:00
|
|
|
nsresult StartupCache::PutBuffer(const char* id, UniquePtr<char[]>&& inbuf,
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
uint32_t len, bool isFromChildProcess) {
|
2010-08-12 23:37:44 +04:00
|
|
|
if (StartupCache::gShutdownInitiated) {
|
|
|
|
return NS_ERROR_NOT_AVAILABLE;
|
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// We use this to update the RequestedByChild flag.
|
|
|
|
MOZ_RELEASE_ASSERT(
|
|
|
|
inbuf || isFromChildProcess,
|
|
|
|
"Null input to PutBuffer should only be used by child actor.");
|
|
|
|
|
|
|
|
if (!XRE_IsParentProcess() && !mChildActor && mFullyInitialized) {
|
|
|
|
return NS_ERROR_NOT_AVAILABLE;
|
|
|
|
}
|
|
|
|
|
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
if (!mLoaded) {
|
|
|
|
return NS_ERROR_NOT_AVAILABLE;
|
|
|
|
}
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
bool exists = mTable.has(nsDependentCString(id));
|
2017-07-06 15:00:35 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
if (exists) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (isFromChildProcess) {
|
|
|
|
decltype(mTable)::Ptr p = mTable.lookup(nsDependentCString(id));
|
|
|
|
auto& value = p->value();
|
|
|
|
value.mFlags += StartupCacheEntryFlags::RequestedByChild;
|
|
|
|
if (value.mRequestedOrder == kStartupCacheEntryNotRequested) {
|
|
|
|
value.mRequestedOrder = ++mRequestedCount;
|
|
|
|
MOZ_ASSERT(mRequestedCount <= mTable.count(),
|
|
|
|
"Somehow we requested more StartupCache items than exist.");
|
|
|
|
ResetStartupWriteTimerCheckingReadCount();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
NS_WARNING("Existing entry in StartupCache.");
|
|
|
|
}
|
|
|
|
|
2015-05-22 23:29:51 +03:00
|
|
|
// Double-caching is undesirable but not an error.
|
|
|
|
return NS_OK;
|
|
|
|
}
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
|
|
|
|
if (!inbuf) {
|
|
|
|
MOZ_ASSERT(false,
|
|
|
|
"Should only PutBuffer with null buffer for existing entries.");
|
|
|
|
return NS_ERROR_UNEXPECTED;
|
|
|
|
}
|
|
|
|
|
|
|
|
EnumSet<StartupCacheEntryFlags> flags =
|
|
|
|
StartupCacheEntryFlags::AddedThisSession;
|
|
|
|
if (isFromChildProcess) {
|
|
|
|
flags += StartupCacheEntryFlags::RequestedByChild;
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
}
|
2015-05-22 23:29:51 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
// putNew returns false on alloc failure - in the very unlikely event we hit
|
|
|
|
// that and aren't going to crash elsewhere, there's no reason we need to
|
|
|
|
// crash here.
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (mTable.putNew(
|
|
|
|
nsCString(id),
|
|
|
|
StartupCacheEntry(std::move(inbuf), len, ++mRequestedCount, flags))) {
|
|
|
|
return ResetStartupWriteTimerImpl();
|
2011-01-07 21:55:14 +03:00
|
|
|
}
|
2019-10-04 23:45:18 +03:00
|
|
|
MOZ_DIAGNOSTIC_ASSERT(mTable.count() < STARTUP_CACHE_MAX_CAPACITY,
|
|
|
|
"Too many StartupCache entries.");
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return NS_OK;
|
2011-10-25 04:50:47 +04:00
|
|
|
}
|
|
|
|
|
2015-07-31 07:19:57 +03:00
|
|
|
size_t StartupCache::HeapSizeOfIncludingThis(
|
|
|
|
mozilla::MallocSizeOf aMallocSizeOf) const {
|
2011-12-19 04:20:36 +04:00
|
|
|
// This function could measure more members, but they haven't been found by
|
|
|
|
// DMD to be significant. They can be added later if necessary.
|
|
|
|
|
2015-07-31 07:19:57 +03:00
|
|
|
size_t n = aMallocSizeOf(this);
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
n += mTable.shallowSizeOfExcludingThis(aMallocSizeOf);
|
2019-10-16 18:41:54 +03:00
|
|
|
for (auto iter = mTable.iter(); !iter.done(); iter.next()) {
|
|
|
|
if (iter.get().value().mData) {
|
|
|
|
n += aMallocSizeOf(iter.get().value().mData.get());
|
|
|
|
}
|
|
|
|
n += iter.get().key().SizeOfExcludingThisIfUnshared(aMallocSizeOf);
|
|
|
|
}
|
2015-07-31 07:19:57 +03:00
|
|
|
|
|
|
|
return n;
|
2011-12-19 04:20:36 +04:00
|
|
|
}
|
|
|
|
|
2011-04-01 09:24:16 +04:00
|
|
|
/**
|
|
|
|
* WriteToDisk writes the cache out to disk. Callers of WriteToDisk need to call
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
* WaitOnWriteComplete to make sure there isn't a write
|
|
|
|
* happening on another thread
|
2011-04-01 09:24:16 +04:00
|
|
|
*/
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
Result<Ok, nsresult> StartupCache::WriteToDisk() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MOZ_ASSERT(XRE_IsParentProcess());
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
if (!mDirty || mWrittenOnce) {
|
|
|
|
return Ok();
|
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
if (!mFile) {
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
nsCOMPtr<nsIFile> tmpFile;
|
|
|
|
nsresult rv = mFile->Clone(getter_AddRefs(tmpFile));
|
|
|
|
if (NS_FAILED(rv)) {
|
|
|
|
return Err(NS_ERROR_UNEXPECTED);
|
|
|
|
}
|
|
|
|
|
|
|
|
nsAutoCString leafName;
|
|
|
|
mFile->GetNativeLeafName(leafName);
|
|
|
|
tmpFile->SetNativeLeafName(leafName + NS_LITERAL_CSTRING("-tmp"));
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
{
|
|
|
|
AutoFDClose fd;
|
|
|
|
MOZ_TRY(tmpFile->OpenNSPRFileDesc(PR_WRONLY | PR_CREATE_FILE | PR_TRUNCATE,
|
|
|
|
0644, &fd.rwget()));
|
|
|
|
|
|
|
|
nsTArray<std::pair<const nsCString*, StartupCacheEntry*>> entries;
|
|
|
|
|
|
|
|
for (auto iter = mTable.iter(); !iter.done(); iter.next()) {
|
2020-03-17 15:42:12 +03:00
|
|
|
entries.AppendElement(
|
|
|
|
std::make_pair(&iter.get().key(), &iter.get().value()));
|
2019-10-04 23:45:18 +03:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (entries.IsEmpty()) {
|
|
|
|
return Ok();
|
|
|
|
}
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
entries.Sort(StartupCacheEntry::Comparator());
|
|
|
|
loader::OutputBuffer buf;
|
|
|
|
for (auto& e : entries) {
|
|
|
|
const auto& key = e.first;
|
|
|
|
auto* value = e.second;
|
|
|
|
auto uncompressedSize = value->mUncompressedSize;
|
|
|
|
uint8_t sharedToChild =
|
|
|
|
value->mFlags.contains(StartupCacheEntryFlags::RequestedByChild);
|
|
|
|
// Set the mHeaderOffsetInFile so we can go back and edit the offset.
|
|
|
|
value->mHeaderOffsetInFile = buf.cursor();
|
|
|
|
// Write a 0 offset/compressed size as a placeholder until we get the real
|
|
|
|
// offset after compressing.
|
|
|
|
buf.codeUint32(0);
|
|
|
|
buf.codeUint32(0);
|
|
|
|
buf.codeUint32(uncompressedSize);
|
|
|
|
buf.codeUint8(sharedToChild);
|
|
|
|
buf.codeString(*key);
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
uint8_t headerSize[4];
|
|
|
|
LittleEndian::writeUint32(headerSize, buf.cursor());
|
|
|
|
|
|
|
|
MOZ_TRY(Write(fd, MAGIC, sizeof(MAGIC)));
|
|
|
|
MOZ_TRY(Write(fd, headerSize, sizeof(headerSize)));
|
|
|
|
size_t prefetchSizeOffset = sizeof(MAGIC) + sizeof(headerSize);
|
|
|
|
size_t headerStart = prefetchSizeOffset + sizeof(uint32_t);
|
|
|
|
size_t dataStart = headerStart + buf.cursor();
|
|
|
|
MOZ_TRY(Seek(fd, dataStart));
|
|
|
|
|
|
|
|
size_t offset = 0;
|
|
|
|
size_t prefetchSize = 0;
|
|
|
|
|
|
|
|
const size_t chunkSize = 1024 * 16;
|
|
|
|
LZ4FrameCompressionContext ctx(6, /* aCompressionLevel */
|
|
|
|
chunkSize, /* aReadBufLen */
|
|
|
|
true, /* aChecksum */
|
|
|
|
true); /* aStableSrc */
|
|
|
|
size_t writeBufLen = ctx.GetRequiredWriteBufferLength();
|
|
|
|
auto writeBuffer = MaybeOwnedCharPtr(writeBufLen);
|
|
|
|
auto writeSpan = MakeSpan(writeBuffer.get(), writeBufLen);
|
|
|
|
|
|
|
|
for (auto& e : entries) {
|
|
|
|
auto* value = e.second;
|
|
|
|
|
|
|
|
// If this is the first entry which was not requested, set the prefetch
|
|
|
|
// size to the offset before we include this item. We will then prefetch
|
|
|
|
// all items up to but not including this one during our next startup.
|
|
|
|
if (value->mRequestedOrder == kStartupCacheEntryNotRequested &&
|
|
|
|
prefetchSize == 0) {
|
|
|
|
prefetchSize = dataStart + offset;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Reuse the existing compressed entry if possible
|
|
|
|
if (mCacheData.initialized() && value->mCompressedSize) {
|
|
|
|
Span<const char> compressed =
|
|
|
|
MakeSpan(mCacheData.get<char>().get() + mCacheEntriesBaseOffset +
|
|
|
|
value->mOffset,
|
|
|
|
value->mCompressedSize);
|
|
|
|
MOZ_TRY(Write(fd, compressed.Elements(), compressed.Length()));
|
|
|
|
value->mOffset = offset;
|
|
|
|
offset += compressed.Length();
|
|
|
|
} else {
|
|
|
|
value->mOffset = offset;
|
|
|
|
Span<const char> result;
|
|
|
|
MOZ_TRY_VAR(result, ctx.BeginCompressing(writeSpan).mapErr(
|
|
|
|
MapLZ4ErrorToNsresult));
|
|
|
|
MOZ_TRY(Write(fd, result.Elements(), result.Length()));
|
|
|
|
offset += result.Length();
|
|
|
|
|
|
|
|
for (size_t i = 0; i < value->mUncompressedSize; i += chunkSize) {
|
|
|
|
size_t size = std::min(chunkSize, value->mUncompressedSize - i);
|
|
|
|
char* uncompressed = value->mData.get() + i;
|
|
|
|
MOZ_TRY_VAR(result,
|
|
|
|
ctx.ContinueCompressing(MakeSpan(uncompressed, size))
|
|
|
|
.mapErr(MapLZ4ErrorToNsresult));
|
|
|
|
MOZ_TRY(Write(fd, result.Elements(), result.Length()));
|
|
|
|
offset += result.Length();
|
|
|
|
}
|
|
|
|
|
|
|
|
MOZ_TRY_VAR(result, ctx.EndCompressing().mapErr(MapLZ4ErrorToNsresult));
|
|
|
|
MOZ_TRY(Write(fd, result.Elements(), result.Length()));
|
|
|
|
offset += result.Length();
|
|
|
|
value->mCompressedSize = offset - value->mOffset;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (prefetchSize == 0) {
|
|
|
|
prefetchSize = dataStart + offset;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (auto& e : entries) {
|
|
|
|
auto* value = e.second;
|
|
|
|
uint8_t* headerEntry = buf.Get() + value->mHeaderOffsetInFile;
|
|
|
|
LittleEndian::writeUint32(headerEntry, value->mOffset);
|
|
|
|
LittleEndian::writeUint32(headerEntry + sizeof(value->mOffset),
|
|
|
|
value->mCompressedSize);
|
|
|
|
}
|
|
|
|
|
|
|
|
uint8_t prefetchSizeBuf[4];
|
|
|
|
LittleEndian::writeUint32(prefetchSizeBuf, prefetchSize);
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MOZ_TRY(Seek(fd, prefetchSizeOffset));
|
|
|
|
MOZ_TRY(Write(fd, prefetchSizeBuf, sizeof(prefetchSizeBuf)));
|
|
|
|
MOZ_TRY(Write(fd, buf.Get(), buf.cursor()));
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
}
|
2011-01-21 00:40:45 +03:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
mDirty = false;
|
|
|
|
mWrittenOnce = true;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mCacheData.reset();
|
|
|
|
tmpFile->MoveToNative(nullptr, leafName);
|
2012-10-11 12:17:15 +04:00
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return Ok();
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
void StartupCache::InvalidateCache() {
|
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
InvalidateCacheImpl();
|
|
|
|
}
|
2020-07-01 22:16:28 +03:00
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
void StartupCache::InvalidateCacheImpl(bool memoryOnly) {
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
|
|
|
|
|
|
|
WaitOnPrefetchThread();
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
mWrittenOnce = false;
|
2019-03-05 19:52:57 +03:00
|
|
|
if (memoryOnly) {
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
// This should only be called in tests.
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
auto writeResult = WriteToDisk();
|
|
|
|
if (NS_WARN_IF(writeResult.isErr())) {
|
|
|
|
gIgnoreDiskCache = true;
|
|
|
|
return;
|
|
|
|
}
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-09-28 01:15:35 +03:00
|
|
|
}
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
|
2020-07-01 22:16:28 +03:00
|
|
|
if (mCurTableReferenced) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
// There should be no way for this assert to fail other than a user
|
|
|
|
// manually sending startupcache-invalidate messages through the Browser
|
|
|
|
// Toolbox.
|
2020-07-01 22:16:28 +03:00
|
|
|
MOZ_DIAGNOSTIC_ASSERT(xpc::IsInAutomation() || mOldTables.Length() < 10,
|
|
|
|
"Startup cache invalidated too many times.");
|
|
|
|
mOldTables.AppendElement(std::move(mTable));
|
|
|
|
mCurTableReferenced = false;
|
|
|
|
} else {
|
|
|
|
mTable.clear();
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-01 20:55:38 +03:00
|
|
|
}
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
|
|
|
|
if (!XRE_IsParentProcess()) {
|
|
|
|
mCacheData.reset();
|
|
|
|
return;
|
|
|
|
}
|
2020-07-01 22:16:28 +03:00
|
|
|
mRequestedCount = 0;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!XRE_IsParentProcess()) {
|
|
|
|
gIgnoreDiskCache = true;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
if (!memoryOnly) {
|
|
|
|
mCacheData.reset();
|
|
|
|
nsresult rv = mFile->Remove(false);
|
|
|
|
if (NS_FAILED(rv) && rv != NS_ERROR_FILE_TARGET_DOES_NOT_EXIST &&
|
|
|
|
rv != NS_ERROR_FILE_NOT_FOUND) {
|
|
|
|
gIgnoreDiskCache = true;
|
|
|
|
return;
|
|
|
|
}
|
2012-10-11 12:17:15 +04:00
|
|
|
}
|
|
|
|
gIgnoreDiskCache = false;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
Unused << mCacheData.init(mFile);
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
auto result = LoadArchive();
|
|
|
|
if (NS_WARN_IF(result.isErr())) {
|
|
|
|
gIgnoreDiskCache = true;
|
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
2020-02-12 22:02:12 +03:00
|
|
|
void StartupCache::MaybeInitShutdownWrite() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!XRE_IsParentProcess()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
if (mWriteTimer) {
|
|
|
|
mWriteTimer->Cancel();
|
2020-02-12 22:02:12 +03:00
|
|
|
}
|
|
|
|
gShutdownInitiated = true;
|
|
|
|
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
MaybeWriteOffMainThread();
|
|
|
|
}
|
|
|
|
|
|
|
|
void StartupCache::EnsureShutdownWriteComplete() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!XRE_IsParentProcess()) {
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
// Otherwise, ensure the write happens. The timer should have been cancelled
|
|
|
|
// already in MaybeInitShutdownWrite.
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!mLock.TryLock()) {
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
// Uh oh, we're writing away from the main thread. Wait to gain the lock,
|
|
|
|
// to ensure the write completes.
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mLock.Lock();
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
} else {
|
|
|
|
// We got the lock. Keep the following in sync with
|
|
|
|
// MaybeWriteOffMainThread:
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
|
|
|
|
// If we've already written or there's nothing to write,
|
|
|
|
// we don't need to do anything. This is the common case.
|
|
|
|
if (mWrittenOnce || (mCacheData.initialized() && !ShouldCompactCache())) {
|
|
|
|
mLock.Unlock();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
WaitOnPrefetchThread();
|
|
|
|
mDirty = true;
|
|
|
|
// Most of this should be redundant given MaybeWriteOffMainThread should
|
|
|
|
// have run before now.
|
|
|
|
|
|
|
|
auto writeResult = WriteToDisk();
|
|
|
|
Unused << NS_WARN_IF(writeResult.isErr());
|
|
|
|
// We've had the lock, and `WriteToDisk()` sets mWrittenOnce and mDirty
|
|
|
|
// when done, and checks for them when starting, so we don't need to do
|
|
|
|
// anything else.
|
|
|
|
}
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mLock.Unlock();
|
2020-02-12 22:02:12 +03:00
|
|
|
}
|
|
|
|
|
2012-10-11 12:17:15 +04:00
|
|
|
void StartupCache::IgnoreDiskCache() {
|
|
|
|
gIgnoreDiskCache = true;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (gStartupCache) {
|
|
|
|
gStartupCache->InvalidateCache();
|
|
|
|
}
|
2012-10-11 12:17:15 +04:00
|
|
|
}
|
|
|
|
|
2019-10-04 23:45:32 +03:00
|
|
|
void StartupCache::WaitOnPrefetchThread() {
|
|
|
|
if (!mPrefetchThread || mPrefetchThread == PR_GetCurrentThread()) return;
|
|
|
|
|
|
|
|
PR_JoinThread(mPrefetchThread);
|
|
|
|
mPrefetchThread = nullptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
void StartupCache::ThreadedPrefetch(void* aClosure) {
|
|
|
|
AUTO_PROFILER_REGISTER_THREAD("StartupCache");
|
|
|
|
NS_SetCurrentThreadName("StartupCache");
|
|
|
|
mozilla::IOInterposer::RegisterCurrentThread();
|
|
|
|
StartupCache* startupCacheObj = static_cast<StartupCache*>(aClosure);
|
2020-04-11 00:16:15 +03:00
|
|
|
uint8_t* buf = startupCacheObj->mCacheData.get<uint8_t>().get();
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
size_t size = startupCacheObj->mPrefetchSize;
|
2020-04-11 00:16:15 +03:00
|
|
|
MMAP_FAULT_HANDLER_BEGIN_BUFFER(buf, size)
|
|
|
|
PrefetchMemory(buf, size);
|
|
|
|
MMAP_FAULT_HANDLER_CATCH()
|
2019-10-04 23:45:32 +03:00
|
|
|
mozilla::IOInterposer::UnregisterCurrentThread();
|
|
|
|
}
|
|
|
|
|
2019-10-04 23:45:18 +03:00
|
|
|
bool StartupCache::ShouldCompactCache() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mLock.AssertCurrentThreadOwns();
|
2019-10-04 23:45:18 +03:00
|
|
|
// If we've requested less than 4/5 of the startup cache, then we should
|
|
|
|
// probably compact it down. This can happen quite easily after the first run,
|
|
|
|
// which seems to request quite a few more things than subsequent runs.
|
|
|
|
CheckedInt<uint32_t> threshold = CheckedInt<uint32_t>(mTable.count()) * 4 / 5;
|
|
|
|
MOZ_RELEASE_ASSERT(threshold.isValid(), "Runaway StartupCache size");
|
|
|
|
return mRequestedCount < threshold.value();
|
|
|
|
}
|
|
|
|
|
2011-01-21 00:40:45 +03:00
|
|
|
/*
|
|
|
|
* The write-thread is spawned on a timeout(which is reset with every write).
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
* This can avoid a slow shutdown.
|
2011-01-21 00:40:45 +03:00
|
|
|
*/
|
2010-08-12 23:37:44 +04:00
|
|
|
void StartupCache::WriteTimeout(nsITimer* aTimer, void* aClosure) {
|
2014-01-11 04:16:25 +04:00
|
|
|
/*
|
|
|
|
* It is safe to use the pointer passed in aClosure to reference the
|
|
|
|
* StartupCache object because the timer's lifetime is tightly coupled to
|
|
|
|
* the lifetime of the StartupCache object; this timer is canceled in the
|
|
|
|
* StartupCache destructor, guaranteeing that this function runs if and only
|
|
|
|
* if the StartupCache object is valid.
|
|
|
|
*/
|
|
|
|
StartupCache* startupCacheObj = static_cast<StartupCache*>(aClosure);
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(startupCacheObj->mLock);
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
startupCacheObj->MaybeWriteOffMainThread();
|
2020-02-12 22:02:12 +03:00
|
|
|
}
|
|
|
|
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
void StartupCache::SendEntriesTimeout(nsITimer* aTimer, void* aClosure) {
|
|
|
|
StartupCache* sc = static_cast<StartupCache*>(aClosure);
|
|
|
|
MutexAutoLock lock(sc->mLock);
|
|
|
|
((StartupCacheChild*)sc->mChildActor)->SendEntriesAndFinalize(sc->mTable);
|
|
|
|
}
|
|
|
|
|
2020-02-12 22:02:12 +03:00
|
|
|
/*
|
|
|
|
* See StartupCache::WriteTimeout above - this is just the non-static body.
|
|
|
|
*/
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
void StartupCache::MaybeWriteOffMainThread() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mLock.AssertCurrentThreadOwns();
|
|
|
|
if (!XRE_IsParentProcess()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
if (mWrittenOnce) {
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2020-02-12 22:02:12 +03:00
|
|
|
if (mCacheData.initialized() && !ShouldCompactCache()) {
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
// Keep this code in sync with EnsureShutdownWriteComplete.
|
2020-02-12 22:02:12 +03:00
|
|
|
WaitOnPrefetchThread();
|
|
|
|
mDirty = true;
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
|
|
|
|
RefPtr<StartupCache> self = this;
|
|
|
|
nsCOMPtr<nsIRunnable> runnable =
|
|
|
|
NS_NewRunnableFunction("StartupCache::Write", [self]() mutable {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(self->mLock);
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
auto result = self->WriteToDisk();
|
|
|
|
Unused << NS_WARN_IF(result.isErr());
|
|
|
|
});
|
|
|
|
NS_DispatchBackgroundTask(runnable.forget(), NS_DISPATCH_EVENT_MAY_BLOCK);
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
// We don't want to refcount StartupCache, so we'll just
|
|
|
|
// hold a ref to this and pass it to observerService instead.
|
2014-04-27 11:06:00 +04:00
|
|
|
NS_IMPL_ISUPPORTS(StartupCacheListener, nsIObserver)
|
2010-08-12 23:37:44 +04:00
|
|
|
|
2014-01-04 19:02:17 +04:00
|
|
|
nsresult StartupCacheListener::Observe(nsISupports* subject, const char* topic,
|
|
|
|
const char16_t* data) {
|
2011-01-21 00:40:45 +03:00
|
|
|
StartupCache* sc = StartupCache::GetSingleton();
|
|
|
|
if (!sc) return NS_OK;
|
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
if (strcmp(topic, NS_XPCOM_SHUTDOWN_OBSERVER_ID) == 0) {
|
2011-01-21 00:40:45 +03:00
|
|
|
// Do not leave the thread running past xpcom shutdown
|
2020-02-12 22:02:12 +03:00
|
|
|
sc->WaitOnPrefetchThread();
|
2011-10-17 18:59:28 +04:00
|
|
|
StartupCache::gShutdownInitiated = true;
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
// Note that we don't do anything special for the background write
|
|
|
|
// task; we expect the threadpool to finish running any tasks already
|
|
|
|
// posted to it prior to shutdown. FastShutdown will call
|
|
|
|
// EnsureShutdownWriteComplete() to ensure any pending writes happen
|
|
|
|
// in that case.
|
2010-09-15 04:39:07 +04:00
|
|
|
} else if (strcmp(topic, "startupcache-invalidate") == 0) {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(sc->mLock);
|
|
|
|
sc->InvalidateCacheImpl(data && nsCRT::strcmp(data, u"memoryOnly") == 0);
|
|
|
|
} else if (sc->mContentStartupFinishedTopic.Equals(topic) &&
|
|
|
|
sc->mChildActor) {
|
|
|
|
if (sProcessType == ProcessType::PrivilegedAbout) {
|
|
|
|
if (!sc->mSendEntriesTimer) {
|
|
|
|
sc->mSendEntriesTimer = NS_NewTimer();
|
|
|
|
sc->mSendEntriesTimer->InitWithNamedFuncCallback(
|
|
|
|
StartupCache::SendEntriesTimeout, sc, 10000,
|
|
|
|
nsITimer::TYPE_ONE_SHOT, "StartupCache::SendEntriesTimeout");
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
MutexAutoLock lock(sc->mLock);
|
|
|
|
((StartupCacheChild*)sc->mChildActor)->SendEntriesAndFinalize(sc->mTable);
|
|
|
|
}
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
2011-01-21 00:40:45 +03:00
|
|
|
return NS_OK;
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
nsresult StartupCache::GetDebugObjectOutputStream(
|
|
|
|
nsIObjectOutputStream* aStream, nsIObjectOutputStream** aOutStream) {
|
|
|
|
NS_ENSURE_ARG_POINTER(aStream);
|
|
|
|
#ifdef DEBUG
|
2016-11-23 15:22:34 +03:00
|
|
|
auto* stream = new StartupCacheDebugOutputStream(aStream, &mWriteObjectMap);
|
2010-08-12 23:37:44 +04:00
|
|
|
NS_ADDREF(*aOutStream = stream);
|
|
|
|
#else
|
|
|
|
NS_ADDREF(*aOutStream = aStream);
|
|
|
|
#endif
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
2019-10-04 23:45:18 +03:00
|
|
|
nsresult StartupCache::ResetStartupWriteTimerCheckingReadCount() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mLock.AssertCurrentThreadOwns();
|
2019-10-04 23:45:18 +03:00
|
|
|
nsresult rv = NS_OK;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!mWriteTimer) {
|
|
|
|
mWriteTimer = NS_NewTimer();
|
|
|
|
} else {
|
|
|
|
rv = mWriteTimer->Cancel();
|
|
|
|
}
|
2019-10-04 23:45:18 +03:00
|
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
2020-06-24 19:00:32 +03:00
|
|
|
// Wait for the specified timeout, then write out the cache.
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mWriteTimer->InitWithNamedFuncCallback(
|
2020-06-24 19:00:32 +03:00
|
|
|
StartupCache::WriteTimeout, this, STARTUP_CACHE_WRITE_TIMEOUT * 1000,
|
|
|
|
nsITimer::TYPE_ONE_SHOT, "StartupCache::WriteTimeout");
|
2019-10-04 23:45:18 +03:00
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
2011-01-07 21:55:14 +03:00
|
|
|
nsresult StartupCache::ResetStartupWriteTimer() {
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
MutexAutoLock lock(mLock);
|
|
|
|
return ResetStartupWriteTimerImpl();
|
|
|
|
}
|
|
|
|
|
|
|
|
nsresult StartupCache::ResetStartupWriteTimerImpl() {
|
|
|
|
if (!XRE_IsParentProcess()) {
|
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
mLock.AssertCurrentThreadOwns();
|
Bug 1550108 - Change StartupCache format from zip to custom r=froydnj
I am not aware of anything that depends on StartupCache being a
zip file, and since I want to use lz4 compression because inflate
is showing up quite a lot in profiles, it's simplest to just use
a custom format. This loosely mimicks the ScriptPreloader code,
with a few diversions:
- Obviously the contents of the cache are compressed. I used lz4
for this as I hit the same file size as deflate at a compression
level of 1, which is what the StartupCache was using previously,
while decompressing an order of magnitude faster. Seemed like
the most conservative change to make. I think it's worth
investigating what the impact of slower algs with higher ratios
would be, but for right now I settled on this. We'd probably
want to look at zstd next.
- I use streaming compression for this via lz4frame. This is not
strictly necessary, but has the benefit of not requiring as
much memory for large buffers, as well as giving us a built-in
checksum, rather than relying on the much slower CRC that we
were doing with the zip-based approach.
- I coded the serialization of the headers inline, since I had to
jump back to add the offset and compressed size, which would
make the nice Code(...) method for the ScriptPreloader stuff
rather more complex. Open to cleaner solutions, but moving it
out just felt like extra hoops for the reader to jump through
to understand without the benefit of being more concise.
Differential Revision: https://phabricator.services.mozilla.com/D34652
--HG--
extra : moz-landing-system : lando
2019-10-04 23:44:59 +03:00
|
|
|
mDirty = true;
|
2017-09-25 05:57:48 +03:00
|
|
|
nsresult rv = NS_OK;
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
if (!mWriteTimer) {
|
|
|
|
mWriteTimer = NS_NewTimer();
|
|
|
|
} else {
|
|
|
|
rv = mWriteTimer->Cancel();
|
|
|
|
}
|
2011-01-07 21:55:14 +03:00
|
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
2020-06-24 19:00:32 +03:00
|
|
|
// Wait for the specified timeout, then write out the cache.
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
mWriteTimer->InitWithNamedFuncCallback(
|
2020-06-24 19:00:32 +03:00
|
|
|
StartupCache::WriteTimeout, this, STARTUP_CACHE_WRITE_TIMEOUT * 1000,
|
|
|
|
nsITimer::TYPE_ONE_SHOT, "StartupCache::WriteTimeout");
|
2011-01-07 21:55:14 +03:00
|
|
|
return NS_OK;
|
|
|
|
}
|
|
|
|
|
Bug 1614795 - use the background task queue for startupcache writes, r=dthayer,decoder
Prior to this patch, the startupcache created its own mWriteThread off which it
wrote to disk. It's initialized by MaybeSpawnWriteThread, which got called
at shutdown, to do the shutdown write if there was any reason to do so, and
from a timer that is re-initialized after every addition to the startup cache,
to run 60s after the last change to the cache.
It then joined that write thread on the main thread (in other words, blocks
on that off-main-thread write completing from the main thread) when:
- xpcom-shutdown fired
- the startupcache itself gets destroyed
- someone calls any of:
* HasEntry
* GetBuffer
* PutBuffer
* InvalidateCache
This patch removes the separate write thread, and instead dispatches a task to
the background task queue, indicating it can block. The task is started in
the same circumstances where we previously used to write (timer from the last
PutBuffer call, and shutdown if necessary).
To ensure it cannot be trying to use the data it writes out (mTable) from
the other thread while that data changes on the main thread, we use a mutex.
The task locks the mutex before starting, and unlocks when finished.
Enumerating the cases that we used to block on joining the thread:
In terms of application shutdown, we expect the background task queue to
either finish the write task, or fail to run it if it hasn't started it yet.
In the FastStartup case, we check if a write was necessary; if so, we
attempt to gain the lock without waiting. If we're successful, the write has
not yet started, and we instead run the write on the main thread. Otherwise,
we retry gaining the lock, blocking this time, thus guaranteeing the
off-the-main-thread write completes.
The task keeps a reference to the startupcache object, so it cannot be
destroyed while the task is pending.
Because the write does not modify `mTable`, and neither does `HasEntry`,
we do not need to do anything there.
In the `GetBuffer` case, we do not modify the table unless we have to read
the entry off disk (memmapped into `mCacheData`). This can only happen if
`mCacheData.initialized()` returns true, and we specifically call
`mCacheData.reset()` before firing off the write task to avoid this.
`mCacheData` is only re-initialized if someone calls `LoadArchive()`,
which can only happen from `Init()` (which is guaranteed not to run
again because this is a singleton), or `InvalidateCache()`, where we lock
the mutex (see below). So this is safe - but we assert on the lock to try
and avoid people breaking this chain of assumptions in the future.
When `PutBuffer` is called, we try to lock the mutex - but if locking fails
(ie the background thread is writing), we simply fail to store the entry
in the startupcache. In practice, this should be rare - it'd happen if
new calls to PutBuffer happen while writing during shutdown (when really,
we don't care) or when it's been 60 seconds since the last PutBuffer so
we started writing the startupcache.
When InvalidateCache is called, we lock the mutex - we shouldn't try to
write while invalidating, or invalidate while writing. This may be slow,
but in practice nothing should call `InvalidateCache` except developer
restarts or the `-purgecaches` commandline flag, so it shouldn't
matter a great deal.
Differential Revision: https://phabricator.services.mozilla.com/D70413
--HG--
extra : moz-landing-system : lando
2020-04-15 23:43:44 +03:00
|
|
|
// Used only in tests:
|
Bug 1627075 - OMT and OMP StartupCache access r=froydnj
The overall goal of this patch is to make the StartupCache accessible anywhere.
There's two main pieces to that equation:
1. Allowing it to be accessed off main thread, which means modifying the
mutex usage to ensure that all data accessed from non-main threads is
protected.
2. Allowing it to be accessed out of the chrome process, which means passing
a handle to a shared cache buffer down to child processes.
Number 1 is somewhat fiddly, but it's all generally straightforward work. I'll
hope that the comments and the code are sufficient to explain what's going on
there.
Number 2 has some decisions to be made:
- The first decision was to pass a handle to a frozen chunk of memory down to
all child processes, rather than passing a handle to an actual file. There's
two reasons for this: 1) since we want to compress the underlying file on
disk, giving that file to child processes would mean they have to decompress
it themselves, eating CPU time. 2) since they would have to decompress it
themselves, they would have to allocate the memory for the decompressed
buffers, meaning they cannot all simply share one big decompressed buffer.
- The drawback of this decision is that we have to load and decompress the
buffer up front, before we spawn any child processes. We attempt to
mitigate this by keeping track of all the entries that child processes
access, and only including those in the frozen decompressed shared buffer.
- We base our implementation of this approach off of the shared preferences
implementation. Hopefully I got all of the pieces to fit together
correctly. They seem to work in local testing and on try, but I think
they require a set of experienced eyes looking carefully at them.
- Another decision was whether to send the handles to the buffers over IPC or
via command line. We went with the command line approach, because the startup
cache would need to be accessed very early on in order to ensure we do not
read from any omnijars, and we could not make that work via IPC.
- Unfortunately this means adding another hard-coded FD, similar to
kPrefMapFileDescriptor. It seems like at the very least we need to rope all
of these together into one place, but I think that should be filed as a
follow-up?
Lastly, because this patch is a bit of a monster to review - first, thank you
for looking at it, and second, the reason we're invested in this is because we
saw a >10% improvement in cold startup times on reference hardware, with a p
value less than 0.01. It's still not abundantly clear how reference hardware
numbers translate to numbers on release, and they certainly don't translate
well to Nightly numbers, but it's enough to convince me that it's worth some
effort.
Depends on D78584
Differential Revision: https://phabricator.services.mozilla.com/D77635
2020-07-02 06:39:46 +03:00
|
|
|
bool StartupCache::StartupWriteComplete() { return !mDirty && mWrittenOnce; }
|
2016-11-11 19:57:08 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
// StartupCacheDebugOutputStream implementation
|
|
|
|
#ifdef DEBUG
|
2014-04-27 11:06:00 +04:00
|
|
|
NS_IMPL_ISUPPORTS(StartupCacheDebugOutputStream, nsIObjectOutputStream,
|
|
|
|
nsIBinaryOutputStream, nsIOutputStream)
|
2010-08-12 23:37:44 +04:00
|
|
|
|
|
|
|
bool StartupCacheDebugOutputStream::CheckReferences(nsISupports* aObject) {
|
|
|
|
nsresult rv;
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
nsCOMPtr<nsIClassInfo> classInfo = do_QueryInterface(aObject);
|
|
|
|
if (!classInfo) {
|
|
|
|
NS_ERROR("aObject must implement nsIClassInfo");
|
2011-10-17 18:59:28 +04:00
|
|
|
return false;
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2012-08-22 19:56:38 +04:00
|
|
|
uint32_t flags;
|
2010-08-12 23:37:44 +04:00
|
|
|
rv = classInfo->GetFlags(&flags);
|
2011-10-17 18:59:28 +04:00
|
|
|
NS_ENSURE_SUCCESS(rv, false);
|
2010-08-12 23:37:44 +04:00
|
|
|
if (flags & nsIClassInfo::SINGLETON) return true;
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
nsISupportsHashKey* key = mObjectMap->GetEntry(aObject);
|
|
|
|
if (key) {
|
|
|
|
NS_ERROR(
|
|
|
|
"non-singleton aObject is referenced multiple times in this"
|
|
|
|
"serialization, we don't support that.");
|
2011-10-17 18:59:28 +04:00
|
|
|
return false;
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
mObjectMap->PutEntry(aObject);
|
2011-10-17 18:59:28 +04:00
|
|
|
return true;
|
2010-08-12 23:37:44 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
// nsIObjectOutputStream implementation
|
2011-09-29 10:19:26 +04:00
|
|
|
nsresult StartupCacheDebugOutputStream::WriteObject(nsISupports* aObject,
|
|
|
|
bool aIsStrongRef) {
|
2010-08-12 23:37:44 +04:00
|
|
|
nsCOMPtr<nsISupports> rootObject(do_QueryInterface(aObject));
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
NS_ASSERTION(rootObject.get() == aObject,
|
|
|
|
"bad call to WriteObject -- call WriteCompoundObject!");
|
2011-09-29 10:19:26 +04:00
|
|
|
bool check = CheckReferences(aObject);
|
2010-08-12 23:37:44 +04:00
|
|
|
NS_ENSURE_TRUE(check, NS_ERROR_FAILURE);
|
|
|
|
return mBinaryStream->WriteObject(aObject, aIsStrongRef);
|
|
|
|
}
|
|
|
|
|
|
|
|
nsresult StartupCacheDebugOutputStream::WriteSingleRefObject(
|
|
|
|
nsISupports* aObject) {
|
|
|
|
nsCOMPtr<nsISupports> rootObject(do_QueryInterface(aObject));
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
NS_ASSERTION(rootObject.get() == aObject,
|
|
|
|
"bad call to WriteSingleRefObject -- call WriteCompoundObject!");
|
2011-09-29 10:19:26 +04:00
|
|
|
bool check = CheckReferences(aObject);
|
2010-08-12 23:37:44 +04:00
|
|
|
NS_ENSURE_TRUE(check, NS_ERROR_FAILURE);
|
|
|
|
return mBinaryStream->WriteSingleRefObject(aObject);
|
|
|
|
}
|
|
|
|
|
|
|
|
nsresult StartupCacheDebugOutputStream::WriteCompoundObject(
|
2011-09-29 10:19:26 +04:00
|
|
|
nsISupports* aObject, const nsIID& aIID, bool aIsStrongRef) {
|
2010-08-12 23:37:44 +04:00
|
|
|
nsCOMPtr<nsISupports> rootObject(do_QueryInterface(aObject));
|
2017-07-06 15:00:35 +03:00
|
|
|
|
2010-08-12 23:37:44 +04:00
|
|
|
nsCOMPtr<nsISupports> roundtrip;
|
|
|
|
rootObject->QueryInterface(aIID, getter_AddRefs(roundtrip));
|
|
|
|
NS_ASSERTION(roundtrip.get() == aObject,
|
|
|
|
"bad aggregation or multiple inheritance detected by call to "
|
|
|
|
"WriteCompoundObject!");
|
|
|
|
|
2011-09-29 10:19:26 +04:00
|
|
|
bool check = CheckReferences(aObject);
|
2010-08-12 23:37:44 +04:00
|
|
|
NS_ENSURE_TRUE(check, NS_ERROR_FAILURE);
|
|
|
|
return mBinaryStream->WriteCompoundObject(aObject, aIID, aIsStrongRef);
|
|
|
|
}
|
|
|
|
|
|
|
|
nsresult StartupCacheDebugOutputStream::WriteID(nsID const& aID) {
|
|
|
|
return mBinaryStream->WriteID(aID);
|
|
|
|
}
|
|
|
|
|
2012-08-22 19:56:38 +04:00
|
|
|
char* StartupCacheDebugOutputStream::GetBuffer(uint32_t aLength,
|
|
|
|
uint32_t aAlignMask) {
|
2010-08-12 23:37:44 +04:00
|
|
|
return mBinaryStream->GetBuffer(aLength, aAlignMask);
|
|
|
|
}
|
|
|
|
|
2012-08-22 19:56:38 +04:00
|
|
|
void StartupCacheDebugOutputStream::PutBuffer(char* aBuffer, uint32_t aLength) {
|
2010-08-12 23:37:44 +04:00
|
|
|
mBinaryStream->PutBuffer(aBuffer, aLength);
|
|
|
|
}
|
|
|
|
#endif // DEBUG
|
|
|
|
|
|
|
|
} // namespace scache
|
|
|
|
} // namespace mozilla
|