dcache: fix quadratic behavior with parallel shrinkers
PATCH 1/1 dcache: fix quadratic behavior with parallel shrinkers
When multiple shrinkers are operating on a directory containing many dentries, it takes much longer than if only one shrinker is operating on the directory.
Call the shrinker instances A and B, which shrink DIR containing NUM dentries.
Assume A wins the race for locking DIR's d_lock, then it goes onto moving all unlinked dentries to its dispose list. When it's done, then B will scan the directory once again, but will find that all dentries are already being shrunk, so it will have an empty dispose list. Both A and B will have found NUM dentries (data.found == NUM).
Now comes the interesting part: A will proceed to shrink the dispose list by killing individual dentries and decrementing the refcount of the parent (which is DIR). NB: decrementing DIR's refcount will block if DIR's d_lock is held. B will shrink a zero size list and then immediately restart scanning the directory, where it will lock DIR's d_lock, scan the remaining dentries and find no dentry to dispose.
So that results in B doing the directory scan over and over again, holding d_lock of DIR, while A is waiting for a chance to decrement refcount of DIR and making very slow progress because of this. B is wasting time and holding up progress of A at the same time.
Proposed fix is to check this situation in B (found some dentries, but all are being shrunk already) and just sleep for some time, before retrying the scan. The sleep is proportional to the number of found dentries.
Test script:
--- 8< --- 8< --- 8< --- 8< --- 8< ---
#!/bin/bash
TESTROOT=/var/tmp/test-root
SUBDIR=$TESTROOT/sub/dir
prepare()
{
rm -rf $TESTROOT
mkdir -p $SUBDIR
for (( i = 0; i < 1000; i++ )); do
for ((j = 0; j < 1000; j++)); do
if test -e $SUBDIR/$i.$j; then
echo "This should not happen!"
exit 1
fi
done
printf "%i (%s) ...\r" $((($i + 1) * $j)) grep dentry /proc/slabinfo | sed -e "s/dentry *\([0-9]*\).*/\1/"
done
}
prepare
printf "\nStarting shrinking\n"
time rmdir $TESTROOT 2> /dev/null
prepare
printf "\nStarting parallel shrinking\n"
time (rmdir $SUBDIR & rmdir $TESTROOT 2> /dev/null & wait)
--- 8< --- 8< --- 8< --- 8< --- 8< ---
--- fs/dcache.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
fs/dcache.c +8 -1
@@ -30,6 +30,7 @@#include <linux/bit_spinlock.h> #include <linux/rculist_bl.h> #include <linux/list_lru.h> +#include <linux/delay.h> #include "internal.h" #include "mount.h"
@@ -1479,9 +1480,15 @@ void shrink_dcache_parent(struct dentry *parent)continue ; } -cond_resched ();if (!data.found)break ; + +/* +* Wait some for other shrinkers to process these found +* dentries. This formula gives about 100ns on average per +* dentry for large number of dentries. +*/ +usleep_range (data.found /15 +1 , data.found /7 +2 ); } }EXPORT_SYMBOL (shrink_dcache_parent);
Signature · 1 line
2.14.3
Al Viro
maintainerQuoted 20 lines · Miklos Szeredi
When multiple shrinkers are operating on a directory containing many dentries, it takes much longer than if only one shrinker is operating on the directory.
Call the shrinker instances A and B, which shrink DIR containing NUM dentries.
Assume A wins the race for locking DIR's
d_lock, then it goes onto moving all unlinked dentries to its dispose list. When it's done, then B will scan the directory once again, but will find that all dentries are already being shrunk, so it will have an empty dispose list. Both A and B will have found NUM dentries (data.found == NUM).Now comes the interesting part: A will proceed to shrink the dispose list by killing individual dentries and decrementing the refcount of the parent (which is DIR). NB: decrementing DIR's refcount will block if DIR's
d_lockis held. B will shrink a zero size list and then immediately restart scanning the directory, where it will lock DIR'sd_lock, scan the remaining dentries and find no dentry to dispose.
Quoted 8 lines
So that results in B doing the directory scan over and over again, holding
d_lockof DIR, while A is waiting for a chance to decrement refcount of DIR and making very slow progress because of this. B is wasting time and holding up progress of A at the same time.Proposed fix is to check this situation in B (found some dentries, but all are being shrunk already) and just sleep for some time, before retrying the scan. The sleep is proportional to the number of found dentries.
The thing is, the majority of massive shrink_dcache_parent() can be killed. Let's do that first and see if anything else is really needed.
As it is, rmdir() and rename() are ridiculously bad - they should only call shrink_dcache_parent() after successful ->rmdir() or ->rename(). Sure, there are other places where we do large shrink_dcache_parent() runs, but those won't trigger in parallel on the same tree.
IOW, let's wait adding complexity until we fix the sources of those calls.
Quoted 30 lines · Al Viro
Quoted 20 lines · Miklos Szeredi
When multiple shrinkers are operating on a directory containing many dentries, it takes much longer than if only one shrinker is operating on the directory.
Call the shrinker instances A and B, which shrink DIR containing NUM dentries.
Assume A wins the race for locking DIR's
d_lock, then it goes onto moving all unlinked dentries to its dispose list. When it's done, then B will scan the directory once again, but will find that all dentries are already being shrunk, so it will have an empty dispose list. Both A and B will have found NUM dentries (data.found == NUM).Now comes the interesting part: A will proceed to shrink the dispose list by killing individual dentries and decrementing the refcount of the parent (which is DIR). NB: decrementing DIR's refcount will block if DIR's
d_lockis held. B will shrink a zero size list and then immediately restart scanning the directory, where it will lock DIR'sd_lock, scan the remaining dentries and find no dentry to dispose.Quoted 8 lines
So that results in B doing the directory scan over and over again, holding
d_lockof DIR, while A is waiting for a chance to decrement refcount of DIR and making very slow progress because of this. B is wasting time and holding up progress of A at the same time.Proposed fix is to check this situation in B (found some dentries, but all are being shrunk already) and just sleep for some time, before retrying the scan. The sleep is proportional to the number of found dentries.
Quoted 7 lines
The thing is, the majority of massive
shrink_dcache_parent()can be killed. Let's do that first and see if anything else is really needed.As it is,
rmdir()andrename()are ridiculously bad - they should only callshrink_dcache_parent()after successful ->rmdir() or ->rename(). Sure, there are other places where we do largeshrink_dcache_parent()runs, but those won't trigger in parallel on the same tree.
I think we are cat hit this also with lru pruner (prune_dcache_sb(), shrink_dcache_sb()) running in parallel with shrink_dcache_parent(). Although shrink_dcache_sb() looks better in this regard, since it will only hold up to 1024 dentries in the dispose list.
I'm open to a better solution, but keep in mind that it will also need backporting.
Thanks,
Miklos
Quoted 39 lines · Miklos Szeredi
Quoted 30 lines · Al Viro
Quoted 20 lines · Miklos Szeredi
When multiple shrinkers are operating on a directory containing many dentries, it takes much longer than if only one shrinker is operating on the directory.
Call the shrinker instances A and B, which shrink DIR containing NUM dentries.
Assume A wins the race for locking DIR's
d_lock, then it goes onto moving all unlinked dentries to its dispose list. When it's done, then B will scan the directory once again, but will find that all dentries are already being shrunk, so it will have an empty dispose list. Both A and B will have found NUM dentries (data.found == NUM).Now comes the interesting part: A will proceed to shrink the dispose list by killing individual dentries and decrementing the refcount of the parent (which is DIR). NB: decrementing DIR's refcount will block if DIR's
d_lockis held. B will shrink a zero size list and then immediately restart scanning the directory, where it will lock DIR'sd_lock, scan the remaining dentries and find no dentry to dispose.Quoted 8 lines
So that results in B doing the directory scan over and over again, holding
d_lockof DIR, while A is waiting for a chance to decrement refcount of DIR and making very slow progress because of this. B is wasting time and holding up progress of A at the same time.Proposed fix is to check this situation in B (found some dentries, but all are being shrunk already) and just sleep for some time, before retrying the scan. The sleep is proportional to the number of found dentries.
Quoted 7 lines
The thing is, the majority of massive
shrink_dcache_parent()can be killed. Let's do that first and see if anything else is really needed.As it is,
rmdir()andrename()are ridiculously bad - they should only callshrink_dcache_parent()after successful ->rmdir() or ->rename(). Sure, there are other places where we do largeshrink_dcache_parent()runs, but those won't trigger in parallel on the same tree.
Quoted 4 lines
I think we are cat hit this also with lru pruner (
prune_dcache_sb(),shrink_dcache_sb()) running in parallel withshrink_dcache_parent(). Althoughshrink_dcache_sb()looks better in this regard, since it will only hold up to 1024 dentries in the dispose list.
Looking more, prune_dcache_sb() will also batch with a max of 1024 objects. Which mitigates the problem, but doesn't make it go away. Killing 1024 dentries still takes on the order of 100us without contention on d_lock. If shrink_dcache_parent() is busy looping on those dentries, then contention will make this much worse.
Thanks,
Miklos