Speed up WB_SYNC_NONE when a WB_SYNC_ALL occurs simultaneously

Page writebacks with WB_SYNC_NONE can take several seconds to complete 
since they wait for the transaction group to close before being 
committed. This is usually not a problem since the caller does not 
need to wait. However, if we're simultaneously doing a writeback 
with WB_SYNC_ALL (e.g via msync), the latter can block for several 
seconds (up to zfs_txg_timeout) due to the active WB_SYNC_NONE 
writeback since it needs to wait for the transaction to complete 
and the PG_writeback bit to be cleared.

This commit deals with 2 cases:

- No page writeback is active. A WB_SYNC_ALL page writeback starts 
  and even completes. But when it's about to check if the PG_writeback 
  bit has been cleared, another writeback with WB_SYNC_NONE starts. 
  The sync page writeback ends up waiting for the non-sync page 
  writeback to complete.

- A page writeback with WB_SYNC_NONE is already active when a 
  WB_SYNC_ALL writeback starts. The WB_SYNC_ALL writeback ends up 
  waiting for the WB_SYNC_NONE writeback.

The fix works by carefully keeping track of active sync/non-sync 
writebacks and committing when beneficial.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Shaan Nobee <sniper111@gmail.com>
Closes #12662
Closes #12790
This commit is contained in:
Shaan Nobee
2022-05-04 00:23:26 +04:00
committed by GitHub
parent a64d757aa4
commit 411f4a018d
17 changed files with 351 additions and 20 deletions
@@ -5,7 +5,8 @@ dist_pkgdata_SCRIPTS = \
mmap_read_001_pos.ksh \
mmap_write_001_pos.ksh \
mmap_libaio_001_pos.ksh \
mmap_seek_001_pos.ksh
mmap_seek_001_pos.ksh \
mmap_sync_001_pos.ksh
dist_pkgdata_DATA = \
mmap.cfg
@@ -0,0 +1,63 @@
#!/bin/ksh -p
#
# This file and its contents are supplied under the terms of the
# Common Development and Distribution License ("CDDL"), version 1.0.
# You may only use this file in accordance with the terms of version
# 1.0 of the CDDL.
#
# A full copy of the text of the CDDL should have accompanied this
# source. A copy of the CDDL is also available via the Internet at
# http://www.illumos.org/license/CDDL.
#
#
# Copyright (c) 2015, 2016 by Delphix. All rights reserved.
#
. $STF_SUITE/include/libtest.shlib
#
# DESCRIPTION:
# msync()s of mmap()'ed file should complete quickly during
# background dirty page writebacks by the kernel.
#
function cleanup
{
log_must eval "echo $saved_vm_dirty_expire_centisecs > /proc/sys/vm/dirty_expire_centisecs"
log_must eval "echo $saved_vm_dirty_background_ratio > /proc/sys/vm/dirty_background_ratio"
log_must eval "echo $saved_vm_dirty_writeback_centisecs > /proc/sys/vm/dirty_writeback_centisecs"
# revert to some sensible defaults if the values we saved
# were incorrect due to a previous run being interrupted
if [ $(</proc/sys/vm/dirty_expire_centisecs) -eq 1 ]; then
log_must eval "echo 3000 > /proc/sys/vm/dirty_expire_centisecs"
fi
if [ $(</proc/sys/vm/dirty_background_ratio) -eq 0 ]; then
log_must eval "echo 10 > /proc/sys/vm/dirty_background_ratio"
fi
if [ $(</proc/sys/vm/dirty_writeback_centisecs) -eq 1 ]; then
log_must eval "echo 500 > /proc/sys/vm/dirty_writeback_centisecs"
fi
}
if ! is_linux; then
log_unsupported "Only supported on Linux, requires /proc/sys/vm/ tunables"
fi
log_onexit cleanup
log_assert "Run the tests for mmap_sync"
read -r saved_vm_dirty_expire_centisecs < /proc/sys/vm/dirty_expire_centisecs
read -r saved_vm_dirty_background_ratio < /proc/sys/vm/dirty_background_ratio
read -r saved_vm_dirty_writeback_centisecs < /proc/sys/vm/dirty_writeback_centisecs
log_must eval "echo 1 > /proc/sys/vm/dirty_expire_centisecs"
log_must eval "echo 1 > /proc/sys/vm/dirty_background_bytes"
log_must eval "echo 1 > /proc/sys/vm/dirty_writeback_centisecs"
log_must mmap_sync
log_pass "mmap_sync tests passed."