Ractor.yield(obj, move: true) and
Ractor.select(..., yield_value: obj, move: true) tried to yield a
value with move semantices, but if the trial is faild, the obj
should not become a moved object.
To keep this rule, `wait_moving` wait status is introduced.
New yield/take process:
(1) If a ractor tried to yield (move:true), make taking racotr's
wait status `wait_moving` and make a moved object by
`ractor_move(obj)` and wakeup taking ractor.
(2) If a ractor tried to take a message from a ractor waiting fo
yielding (move:true), wakeup the ractor and wait for (1).
When `literal`, check if the literal about to be assigned to a
constant is ractor-shareable, otherwise raise `Ractor::Error` at
runtime instead of `SyntaxError`.
separate some fields from rb_ractor_t to rb_ractor_pub and put it
at the beggining of rb_ractor_t and declare it in vm_core.h so
vm_core.h can access rb_ractor_pub fields.
Now rb_ec_ractor_hooks() is a complete inline function and no
MJIT related issue.
Ractor has several restrictions to keep each ractor being isolated
and some operation such as `CONST="foo"` in non-main ractor raises
an exception. This kind of operation raises an error but there is
confusion (some code raises RuntimeError and some code raises
NameError).
To make clear we introduce Ractor::IsolationError which is raised
when the isolation between ractors is violated.
Ractor.make_shareable(obj) tries to make obj a shareable object
by changing the attribute of obj and traversable objects from obj
(mainly freeze them).
"copy: true" option is more conservative approach by make deep
copied object and make it sharable. It doesn't affect any existing
objects.
This adds rb_category_compile_warn in order to emit compiler warnings
with categories. Note that Ripper currently ignores the category
for these warnings, but by default it ignores the warnings completely,
so this shouldn't matter.
ractor_sleep() can remain wait.status by interrupt, so that this
patch handles more correctly.
This patch fixed this kind of assertion failures:
Assertion Failed: ../src/ractor.c:1332:ractor_yield_atexit:cr->sync.wait.status == wait_none
Thread's interrupt set Ractor's wakeup_status as interrupted, but
the status remains next Ractor communication API. This patch makes
to ignore the previous interrupt state.
[Bug #17366]
Also this patch solves the Thread#kill and Ractor#take issues.
On windows, MJIT doesn't work without this patch because of
the declaration of ruby_single_main_ractor. This patch fix this
issue and move the definition of it from ractor.c to vm.c to locate
near place of ruby_current_vm_ptr.
Now object allocation requires VM global lock to synchronize objspace.
However, of course, it introduces huge overhead.
This patch caches some slots (in a page) by each ractor and use cached
slots for object allocation. If there is no cached slots, acquire the global lock
and get new cached slots, or start GC (marking or lazy sweeping).
accessing theap needs complicating synchronization but it reduce
performance on multi-ractor mode. So simply stop using theap
on multi-ractor mode. In future, theap should be replaced with
more cleaver memory strategy.
ruby_multi_ractor was a flag that indicates the interpreter doesn't
make any additional ractors (single ractor mode).
Instead of boolean flag, ruby_single_main_ractor pointer is introduced
which keeps main ractor's pointer if single ractor mode. If additional
ractors are created, ruby_single_main_ractor becomes NULL.
C extensions can violate the ractor-safety, so only ractor-safe
C extensions (C methods) can run on non-main ractors.
rb_ext_ractor_safe(true) declares that the successive
defined methods are ractor-safe. Otherwiwze, defined methods
checked they are invoked in main ractor and raise an error
if invoked at non-main ractors.
[Feature #17307]
ractor_copy() used rb_ary_modify() to make sure this array is not
sharing anything, but it also checks frozen flag. So frozen arrays
raises an error. To solve this issue, this patch introduces new
function rb_ary_cancel_sharing() which makes sure the array does not
share another array and it doesn't check frozen flag.
[Bug #17343]
A test is quoted from https://github.com/ruby/ruby/pull/3817
To manage ractor-local data for C extension, the following APIs
are defined.
* rb_ractor_local_storage_value_newkey
* rb_ractor_local_storage_value
* rb_ractor_local_storage_value_set
* rb_ractor_local_storage_ptr_newkey
* rb_ractor_local_storage_ptr
* rb_ractor_local_storage_ptr_set
At first, you need to create a key of storage by
rb_ractor_local_(value|ptr)_newkey().
For ptr storage, it accepts the type of storage,
how to mark and how to free with ractor's lifetime.
rb_ractor_local_storage_value/set are used to access a VALUE
and rb_ractor_local_storage_ptr/set are used to access a pointer.
random.c uses this API.
Random generators are not Ractor-safe, so we need to prepare
per-ractor default random genearators. This patch set
`Random::DEFAULT = Randm` (not a Random instance, but the Random
class) and singleton methods like `Random.rand()` use a per-ractor
random generator.
[Feature #17322]
To make some kind of Ractor related extensions, some functions
should be exposed.
* include/ruby/thread_native.h
* rb_native_mutex_*
* rb_native_cond_*
* include/ruby/ractor.h
* RB_OBJ_SHAREABLE_P(obj)
* rb_ractor_shareable_p(obj)
* rb_ractor_std*()
* rb_cRactor
and rm ractor_pub.h
and rename srcdir/ractor.h to srcdir/ractor_core.h
(to avoid conflict with include/ruby/ractor.h)
T_DATA objects can refer unshareable objects and they should be
copied recursively, however there is no way to replace with copied
unshareable objects. However, if a T_DATA object refers only
shareable objects, there is no need to replace. So this kind of
T_DATA object (such as Time, Dir, File::Status and so on) can be
sent by Ractor.send.
`r = Ractor.new{ expr }` generates the block return value from `expr`
and we can get this value by `r.take`. Ractor.yield and Ractor#take
passing values by copying on default. However, the block return value
(we named it "will" in the code) is not referred from the Ractor
because the Ractor is already dead. So we can pass the reference
of "will" to another ractor without copying. We can apply same story
for the propagated exception.
Ractor.make_shareable() supports Proc object if
(1) a Proc only read outer local variables (no assignments)
(2) read outer local variables are shareable.
Read local variables are stored in a snapshot, so after making
shareable Proc, any assignments are not affeect like that:
```ruby
a = 1
pr = Ractor.make_shareable(Proc.new{p a})
pr.call #=> 1
a = 2
pr.call #=> 1 # `a = 2` doesn't affect
```
[Feature #17284]
Introduce new method Ractor.make_shareable(obj) which tries to make
obj shareable object. Protocol is here.
(1) If obj is shareable, it is shareable.
(2) If obj is not a shareable object and if obj can be shareable
object if it is frozen, then freeze obj. If obj has reachable
objects (rs), do rs.each{|o| Ractor.make_shareable(o)}
recursively (recursion is not Ruby-level, but C-level).
(3) Otherwise, raise Ractor::Error. Now T_DATA is not a shareable
object even if the object is frozen.
If the method finished without error, given obj is marked as
a sharable object.
To allow makng a shareable frozen T_DATA object, then set
`RUBY_TYPED_FROZEN_SHAREABLE` as type->flags. On default,
this flag is not set. It means user defined T_DATA objects are
not allowed to become shareable objects when it is frozen.
You can make any object shareable by setting FL_SHAREABLE flag,
so if you know that the T_DATA object is shareable (== thread-safe),
set this flag, at creation time for example. `Ractor` object is one
example, which is not a frozen, but a shareable object.
Ractor#close_outgoing should cancel waiting Ractor.yield. However,
yield a value by the Ractor's block should not cancel (to recognize
terminating Ractor, introduce rb_ractor_t::yield_atexit flag).
rb_ractor_main_p() need to access to the ractor pointer in TLS.
However it is slow operation so that we need to skip this check
if it is not multi-ractor mode (!ruby_multi_ractor).
This performance regression is pointed at
https://bugs.ruby-lang.org/issues/17100#note-27
This commit introduces Ractor mechanism to run Ruby program in
parallel. See doc/ractor.md for more details about Ractor.
See ticket [Feature #17100] to see the implementation details
and discussions.
[Feature #17100]
This commit does not complete the implementation. You can find
many bugs on using Ractor. Also the specification will be changed
so that this feature is experimental. You will see a warning when
you make the first Ractor with `Ractor.new`.
I hope this feature can help programmers from thread-safety issues.
Koichi Sasada
Nobuyoshi Nakada
Marc-Andre Lafortune
Kazuhiro NISHIYAMA
Victor Shepelev
Jeremy Evans
Takashi Kokubun
tompng
Benoit Daloze
Quang-Minh Nguyen