WSL2-Linux-Kernel/fs/xfs
Dave Chinner 80168676eb xfs: force background CIL push under sustained load
I have been seeing occasional pauses in transaction throughput up to
30s long under heavy parallel workloads. The only notable thing was
that the xfsaild was trying to be active during the pauses, but
making no progress. It was running exactly 20 times a second (on the
50ms no-progress backoff), and the number of pushbuf events was
constant across this time as well.  IOWs, the xfsaild appeared to be
stuck on buffers that it could not push out.

Further investigation indicated that it was trying to push out inode
buffers that were pinned and/or locked. The xfsbufd was also getting
woken at the same frequency (by the xfsaild, no doubt) to push out
delayed write buffers. The xfsbufd was not making any progress
because all the buffers in the delwri queue were pinned. This scan-
and-make-no-progress dance went one in the trace for some seconds,
before the xfssyncd came along an issued a log force, and then
things started going again.

However, I noticed something strange about the log force - there
were way too many IO's issued. 516 log buffers were written, to be
exact. That added up to 129MB of log IO, which got me very
interested because it's almost exactly 25% of the size of the log.
He delayed logging code is suppose to aggregate the minimum of 25%
of the log or 8MB worth of changes before flushing. That's what
really puzzled me - why did a log force write 129MB instead of only
8MB?

Essentially what has happened is that no CIL pushes had occurred
since the previous tail push which cleared out 25% of the log space.
That caused all the new transactions to block because there wasn't
log space for them, but they kick the xfsaild to push the tail.
However, the xfsaild was not making progress because there were
buffers it could not lock and flush, and the xfsbufd could not flush
them because they were pinned. As a result, both the xfsaild and the
xfsbufd could not move the tail of the log forward without the CIL
first committing.

The cause of the problem was that the background CIL push, which
should happen when 8MB of aggregated changes have been committed, is
being held off by the concurrent transaction commit load. The
background push does a down_write_trylock() which will fail if there
is a concurrent transaction commit holding the push lock in read
mode. With 8 CPUs all doing transactions as fast as they can, there
was enough concurrent transaction commits to hold off the background
push until tail-pushing could no longer free log space, and the halt
would occur.

It should be noted that there is no reason why it would halt at 25%
of log space used by a single CIL checkpoint. This bug could
definitely violate the "no transaction should be larger than half
the log" requirement and hence result in corruption if the system
crashed under heavy load. This sort of bug is exactly the reason why
delayed logging was tagged as experimental....

The fix is to start blocking background pushes once the threshold
has been exceeded. Rework the threshold calculations to keep the
amount of log space a CIL checkpoint can use to below that of the
AIL push threshold to avoid the problem completely.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-09-29 07:51:03 -05:00
..
linux-2.6 xfs: log IO completion workqueue is a high priority queue 2010-09-10 10:16:54 -05:00
quota Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs-2.6 2010-08-07 12:57:07 -07:00
support xfs: drop dmapi hooks 2010-07-26 13:16:33 -05:00
Kconfig xfs: use generic Posix ACL code 2009-06-10 17:07:47 +02:00
Makefile xfs: simplify log item descriptor tracking 2010-07-26 13:16:34 -05:00
xfs.h xfs: event tracing support 2009-12-14 23:08:16 -06:00
xfs_acl.h xfs: constify xattr_handler 2010-05-21 18:31:19 -04:00
xfs_ag.h xfs: fix access to upper inodes without inode64 2010-05-28 15:19:56 -05:00
xfs_alloc.c xfs: fix gcc 4.6 set but not read and unused statement warnings 2010-07-26 13:16:51 -05:00
xfs_alloc.h xfs: do not use emums for flags used in tracing 2010-07-26 13:16:43 -05:00
xfs_alloc_btree.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_alloc_btree.h [XFS] Always use struct xfs_btree_block instead of short / longform 2008-10-30 17:14:34 +11:00
xfs_arch.h xfs: use generic Posix ACL code 2009-06-10 17:07:47 +02:00
xfs_attr.c xfs: remove unused delta tracking code in xfs_bmapi 2010-07-26 13:16:39 -05:00
xfs_attr.h xfs: convert attr to use unsigned names 2010-01-20 10:47:48 +11:00
xfs_attr_leaf.c xfs: remove unused delta tracking code in xfs_bmapi 2010-07-26 13:16:39 -05:00
xfs_attr_leaf.h [XFS] Remove macro-to-function indirections in attr code 2009-01-09 15:46:44 +11:00
xfs_attr_sf.h xfs: convert attr to use unsigned names 2010-01-20 10:47:48 +11:00
xfs_bit.c [XFS] Use the generic bitops rather than implementing them ourselves. 2008-08-13 15:41:12 +10:00
xfs_bit.h [XFS] Remove macro-to-function indirections in the mask code 2009-01-09 15:53:54 +11:00
xfs_bmap.c xfs: Make fiemap work with sparse files 2010-09-03 09:02:11 -05:00
xfs_bmap.h xfs: remove unused delta tracking code in xfs_bmapi 2010-07-26 13:16:39 -05:00
xfs_bmap_btree.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_bmap_btree.h xfs: make several more functions static 2010-01-15 15:31:38 -06:00
xfs_btree.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_btree.h xfs: add more statics & drop some unused functions 2009-08-31 14:46:20 -05:00
xfs_btree_trace.c [XFS] make btree tracing generic 2008-10-30 16:58:50 +11:00
xfs_btree_trace.h xfs: event tracing support 2009-12-14 23:08:16 -06:00
xfs_buf_item.c xfs: kill the b_strat callback in xfs_buf 2010-07-26 13:16:52 -05:00
xfs_buf_item.h xfs: give li_cb callbacks the correct prototype 2010-07-26 13:16:35 -05:00
xfs_da_btree.c xfs: fix gcc 4.6 set but not read and unused statement warnings 2010-07-26 13:16:51 -05:00
xfs_da_btree.h xfs: convert dirnameops to unsigned char names 2010-01-20 10:47:17 +11:00
xfs_dfrag.c xfs: simplify inode to transaction joining 2010-07-26 13:16:36 -05:00
xfs_dfrag.h xfs: clean up inconsistent variable naming in xfs_swap_extent 2010-01-15 15:31:23 -06:00
xfs_dinode.h xfs: remove m_litino 2009-03-29 09:51:14 +02:00
xfs_dir2.c xfs: split xfs_itrace_entry 2010-07-26 13:16:44 -05:00
xfs_dir2.h xfs: make xfs_dir_cilookup_result use unsigned char 2010-01-20 10:47:25 +11:00
xfs_dir2_block.c xfs: fix gcc 4.6 set but not read and unused statement warnings 2010-07-26 13:16:51 -05:00
xfs_dir2_block.h
xfs_dir2_data.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_dir2_data.h xfs: fix various typos 2009-03-29 09:55:42 +02:00
xfs_dir2_leaf.c xfs: remove unused delta tracking code in xfs_bmapi 2010-07-26 13:16:39 -05:00
xfs_dir2_leaf.h
xfs_dir2_node.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_dir2_node.h xfs: make several more functions static 2010-01-15 15:31:38 -06:00
xfs_dir2_sf.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_dir2_sf.h [XFS] kill xfs_dinode_core_t 2008-12-01 11:37:35 +11:00
xfs_error.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_error.h xfs: add const qualifiers to xfs error function args 2010-05-19 09:58:11 -05:00
xfs_extfree_item.c xfs: fix the xfs_log_iovec i_addr type 2010-07-26 13:16:36 -05:00
xfs_extfree_item.h [XFS] remove always-true #ifndef HAVE_FORMAT32 tests 2009-01-22 14:07:31 +11:00
xfs_filestream.c xfs: clean up filestreams helpers 2010-07-26 13:16:51 -05:00
xfs_filestream.h xfs: clean up filestreams helpers 2010-07-26 13:16:51 -05:00
xfs_fs.h xfs: Make fiemap work with sparse files 2010-09-03 09:02:11 -05:00
xfs_fsops.c xfs: dummy transactions should not dirty VFS state 2010-08-24 11:46:31 +10:00
xfs_fsops.h xfs: dummy transactions should not dirty VFS state 2010-08-24 11:46:31 +10:00
xfs_ialloc.c xfs: fix untrusted inode number lookup 2010-08-24 11:42:30 +10:00
xfs_ialloc.h xfs: rationalize xfs_inobt_lookup* 2009-09-01 12:45:39 -05:00
xfs_ialloc_btree.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_ialloc_btree.h xfs: remove superflous inobt macros 2009-02-09 08:37:14 +01:00
xfs_iget.c xfs: fix gcc 4.6 set but not read and unused statement warnings 2010-07-26 13:16:51 -05:00
xfs_inode.c xfs: ensure we mark all inodes in a freed cluster XFS_ISTALE 2010-08-24 11:42:41 +10:00
xfs_inode.h xfs: simplify and remove xfs_ireclaim 2010-07-26 13:16:48 -05:00
xfs_inode_item.c xfs: fix big endian build 2010-07-26 16:07:38 -05:00
xfs_inode_item.h xfs: simplify inode to transaction joining 2010-07-26 13:16:36 -05:00
xfs_inum.h xfs: remove XFS_INO64_OFFSET 2009-08-31 14:46:22 -05:00
xfs_iomap.c xfs: small cleanups for xfs_iomap / __xfs_get_blocks 2010-07-26 13:16:42 -05:00
xfs_iomap.h xfs: do not use emums for flags used in tracing 2010-07-26 13:16:43 -05:00
xfs_itable.c xfs: remove xfs_iput 2010-07-26 13:16:44 -05:00
xfs_itable.h xfs: remove block number from inode lookup code 2010-06-24 11:35:17 +10:00
xfs_log.c xfs: Reduce log force overhead for delayed logging 2010-08-24 11:40:03 +10:00
xfs_log.h xfs: remove the unused XFS_LOG_SLEEP and XFS_LOG_NOSLEEP flags 2010-07-26 13:16:38 -05:00
xfs_log_cil.c xfs: force background CIL push under sustained load 2010-09-29 07:51:03 -05:00
xfs_log_priv.h xfs: force background CIL push under sustained load 2010-09-29 07:51:03 -05:00
xfs_log_recover.c xfs: fix the xfs_log_iovec i_addr type 2010-07-26 13:16:36 -05:00
xfs_log_recover.h xfs: Clean up XFS_BLI_* flag namespace 2010-05-24 10:33:39 -05:00
xfs_mount.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_mount.h xfs: remove obsolete osyncisosync mount option 2010-07-26 13:16:51 -05:00
xfs_mru_cache.c xfs: Kill filestreams cache flush 2010-01-15 15:34:22 -06:00
xfs_mru_cache.h xfs: Kill filestreams cache flush 2010-01-15 15:34:22 -06:00
xfs_quota.h xfs: removed unused XFS_QMOPT_ flags 2010-05-19 09:58:15 -05:00
xfs_refcache.h
xfs_rename.c xfs: split xfs_itrace_entry 2010-07-26 13:16:44 -05:00
xfs_rtalloc.c xfs: remove unused delta tracking code in xfs_bmapi 2010-07-26 13:16:39 -05:00
xfs_rtalloc.h xfs: be more explicit if RT mount fails due to config 2010-05-28 14:58:24 -05:00
xfs_rw.c xfs: remove unneeded #include statements 2010-07-26 13:16:33 -05:00
xfs_rw.h xfs: only clear the suid bit once in xfs_write 2010-02-12 13:43:57 -06:00
xfs_sb.h [XFS] Remove the rest of the macro-to-function indirections. 2009-01-19 14:45:55 +11:00
xfs_trans.c xfs: unlock items before allowing the CIL to commit 2010-08-24 11:42:52 +10:00
xfs_trans.h xfs: remove the unused XFS_TRANS_NOSLEEP/XFS_TRANS_WAIT flags 2010-07-26 13:16:38 -05:00
xfs_trans_ail.c xfs: drop dmapi hooks 2010-07-26 13:16:33 -05:00
xfs_trans_buf.c xfs: give li_cb callbacks the correct prototype 2010-07-26 13:16:35 -05:00
xfs_trans_extfree.c xfs: simplify log item descriptor tracking 2010-07-26 13:16:34 -05:00
xfs_trans_inode.c xfs: simplify inode to transaction joining 2010-07-26 13:16:36 -05:00
xfs_trans_priv.h xfs: unlock items before allowing the CIL to commit 2010-08-24 11:42:52 +10:00
xfs_trans_space.h xfs: remove superflous inobt macros 2009-02-09 08:37:14 +01:00
xfs_types.h xfs: make the log ticket ID available outside the log infrastructure 2010-05-24 10:33:52 -05:00
xfs_utils.c xfs: simplify xfs_truncate_file 2010-07-26 13:16:52 -05:00
xfs_utils.h xfs: simplify xfs_truncate_file 2010-07-26 13:16:52 -05:00
xfs_vnodeops.c xfs: prevent 32bit overflow in space reservation 2010-09-03 12:19:33 +10:00
xfs_vnodeops.h xfs: kill xfs_lrw.h 2010-03-01 16:35:44 -06:00