btrfs: Test if btrfs will commit too many transaction for balance
PATCH v2 1/1 btrfs: Test if btrfs will commit too many transaction for balance
Kernel commit 64403612b73a ("btrfs: rework btrfs_check_space_for_delayed_refs") is introducing a regression for btrfs balance performance.
Since that commit will cause btrfs to commit transaction for nothing during transaction, it will slow balance dramatically even we only need to relocate several megabytes.
This test case will catch the problem by using super block generation as failure criteria.
For small chunk relocated, we will commit 6 transaction for each block group, and the test case should only have 2 block groups, it should only commit 12 transactions.
This test case will use 120 as the threshold to detect the failure.
And in my test environment, with kernel fix btrfs committed 14 transactions and finished in 6s.
While without the fix btrfs committed 1734 transactions and takes 25s to finish.
So the test case should be more than enough to detect the regression, while still keep the runtime small enough for failure.
---
changelog:
v2:
- Remove the unnecessary workaround for ENOSPC balance Since we know the cause of that false ENOSPC error, a simple sync would be enough to take care of the problem
- Remove unnecessary mount option of max_inline
The same reason above.
--- tests/btrfs/181 | 101 ++++++++++++++++++++++++++++++++++++++++++++ tests/btrfs/181.out | 2 + tests/btrfs/group | 1 + 3 files changed, 104 insertions(+) create mode 100755 tests/btrfs/181 create mode 100644 tests/btrfs/181.out
tests/btrfs/181 new file +101 -0
@@ -0,0 +1,101 @@ +#! /bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (C) 2019 SUSE Linux Products GmbH. All Rights Reserved. +# +# FS QA Test 181 +# +# Test if btrfs will commit too many transactions for nothing and cause +# performance regression during balance. +# +# This bug is going to be fixed by a patch for kenerl title +# "btrfs: don't end the transaction for delayed refs in throttle" +# +seq=`basename $0` +seqres=$RESULT_DIR/$seq +echo "QA output created by $seq" + +here=`pwd` +tmp=/tmp/$$ +status=1 # failure is the default! +trap "_cleanup; exit \$status" 0 1 2 3 15 + +_cleanup() +{ + cd / + rm -f $tmp.* +} + +# get standard environment, filters and checks +. ./common/rc +. ./common/filter + +# remove previous $seqres.full before test +rm -f $seqres.full + +# real QA test starts here + +# Modify as appropriate. +_supported_fs btrfs +_supported_os Linux +_require_scratch +_require_btrfs_command inspect-internal dump-super + +_scratch_mkfs > /dev/null + +_scratch_mount + +nr_files=1024 +nr_snapshots=8 + +get_super_gen() +{ + $BTRFS_UTIL_PROG inspect dump-super "$SCRATCH_DEV" | grep ^generation |\ + awk '{print $2}' +} + +$BTRFS_UTIL_PROG subv create "$SCRATCH_MNT/subvol" > /dev/null + +# Create some small files to take up enough metadata reserved space +for ((i = 0; i < $nr_files; i++)) do + _pwrite_byte 0xcd 0 1K "$SCRATCH_MNT/subvol/file_$i" > /dev/null +done + +# Commit the fs so we can get a stable super generation +sync + +before_gen=$(get_super_gen) +if [ -z $before_gen ]; then + _fail "failed to get super block genreation" +fi + +$BTRFS_UTIL_PROG balance start -m "$SCRATCH_MNT" > /dev/null + +after_gen=$(get_super_gen) + +# Since the fs is pretty small, we should have only 1 small metadata chunk and +# one tiny system chunk. +# Relocating such small chunks only need 6 commits for each, thus 12 commits for +# 2 chunks. +# Here we use 10x the theoretic value as threshold. +theoretic_gen=$(( 6 * 2 )) +threshold_gen=$(( 10 * $theoretic_gen )) +if [ $(( $after_gen - $before_gen )) -gt 120 ]; then + echo "balance committed too many transactions" + echo "super generation before balance: ${before_gen}" + echo "super generation after balance: ${after_gen}" + echo "super generation difference: $((after_gen - before_gen))" + echo "theoretic generation difference: ${theoretic_gen}" + echo "threshold generation difference: ${threshold_gen}" +fi + +echo "super generation before balance: ${before_gen}" >> $seqres.full +echo "super generation after balance: ${after_gen}" >> $seqres.full +echo "super generation difference: $((after_gen - before_gen))" >> $seqres.full +echo "theoretic generation difference: ${theoretic_gen}" >> $seqres.full +echo "threshold generation difference: ${threshold_gen}" >> $seqres.full + +# success, all done +echo "Silence is golden" + +status=0 +exit
tests/btrfs/181.out new file +2 -0
@@ -0,0 +1,2 @@ +QA output created by 181 +Silence is golden
tests/btrfs/group +1 -0
@@ -183,3 +183,4 @@ 178 auto quick send 179 auto qgroup dangerous 180 auto quick qgroup limit +181 auto quick balance
Signature · 1 line
2.18.0
Quoted 7 lines · Qu Wenruo
Kernel commit
64403612b73a("btrfs: reworkbtrfs_check_space_for_delayed_refs") is introducing a regression for btrfs balance performance.Since that commit will cause btrfs to commit transaction for nothing during transaction,
to commit transaction for nothing during transaction -> to commit too many transactions for nothing during balance/relocation.
Also, for the subject, "transaction" -> "transactions" (plural).
Quoted 43 lines
it will slow balance dramatically even we only need to relocate several megabytes.
This test case will catch the problem by using super block generation as failure criteria.
For small chunk relocated, we will commit 6 transaction for each block group, and the test case should only have 2 block groups, it should only commit 12 transactions.This test case will use 120 as the threshold to detect the failure.
And in my test environment, with kernel fix btrfs committed 14 transactions and finished in 6s.
While without the fix btrfs committed 1734 transactions and takes 25s to finish.So the test case should be more than enough to detect the regression, while still keep the runtime small enough for failure.
Signed-off-by Qu Wenruo <[email protected]>---
changelog:
v2:
- Remove the unnecessary workaround for ENOSPC balance Since we know the cause of that false ENOSPC error, a simple sync would be enough to take care of the problem- Remove unnecessary mount option of
max_inline
The same reason above.--- tests/btrfs/181 | 101 ++++++++++++++++++++++++++++++++++++++++++++ tests/btrfs/181.out | 2 + tests/btrfs/group | 1 + 3 files changed, 104 insertions(+) create mode 100755 tests/btrfs/181 create mode 100644 tests/btrfs/181.outtests/btrfs/181 new file +0
-0
tests/btrfs/181 line 10
Quoted 10 lines
tests/btrfs/181 new file +10
-0@@ -0,0 +1,101 @@ +#! /bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (C) 2019 SUSE Linux Products GmbH. All Rights Reserved. +# +# FS QA Test 181 +# +# Test if btrfs will commit too many transactions for nothing and cause +# performance regression during balance. +# +# This bug is going to be fixed by a patch for kenerl title
kenerl -> kernel
Quoted 28 lines
tests/btrfs/181 new file +28
-0+# "btrfs: don't end the transaction for delayed refs in throttle" +# +seq=`basename $0` +seqres=$RESULT_DIR/$seq +echo "QA output created by $seq" + +here=`pwd` +tmp=/tmp/$$ +status=1 # failure is the default! +trap "_cleanup; exit \$status" 0 1 2 3 15 + +_cleanup() +{ + cd / + rm -f $tmp.* +} + +# get standard environment, filters and checks +. ./common/rc +. ./common/filter + +# remove previous $seqres.full before test +rm -f $seqres.full + +# real QA test starts here + +# Modify as appropriate. +_supported_fs btrfs
tests/btrfs/181 line 48
Quoted 10 lines
tests/btrfs/181 new file +10
-0+_supported_os Linux +_require_scratch +_require_btrfs_command inspect-internal dump-super + +_scratch_mkfs > /dev/null + +_scratch_mount + +nr_files=1024 +nr_snapshots=8
nr_snapshots is unused in the test.
tests/btrfs/181 line 56
Quoted 8 lines
tests/btrfs/181 new file +8
-0+ +get_super_gen() +{ + $BTRFS_UTIL_PROG inspect dump-super "$SCRATCH_DEV" | grep ^generation |\ + awk '{print $2}' +} + +$BTRFS_UTIL_PROG subv create "$SCRATCH_MNT/subvol" > /dev/null
subv -> subvolume
We always prefer to use full command names.
tests/btrfs/181 line 68
Quoted 12 lines
tests/btrfs/181 new file +12
-0+ +# Create some small files to take up enough metadata reserved space +for ((i = 0; i < $nr_files; i++)) do + _pwrite_byte 0xcd 0 1K "$SCRATCH_MNT/subvol/file_$i" > /dev/null +done + +# Commit the fs so we can get a stable super generation +sync + +before_gen=$(get_super_gen) +if [ -z $before_gen ]; then + _fail "failed to get super block genreation"
genreation -> generation
tests/btrfs/181 line 69
Quoted 1 line
tests/btrfs/181 new file +1
-0+fi
Also, this error checking should fit better in the get_super_gen function. We check for the error here but not when getting "after_gen", which seems odd.
tests/btrfs/181 line 77
Quoted 8 lines
tests/btrfs/181 new file +8
-0+ +$BTRFS_UTIL_PROG balance start -m "$SCRATCH_MNT" > /dev/null + +after_gen=$(get_super_gen) + +# Since the fs is pretty small, we should have only 1 small metadata chunk and +# one tiny system chunk. +# Relocating such small chunks only need 6 commits for each, thus 12 commits for
need -> needs
Quoted 11 lines
tests/btrfs/181 new file +11
-0+# 2 chunks. +# Here we use 10x the theoretic value as threshold. +theoretic_gen=$(( 6 * 2 )) +threshold_gen=$(( 10 * $theoretic_gen )) +if [ $(( $after_gen - $before_gen )) -gt 120 ]; then + echo "balance committed too many transactions" + echo "super generation before balance: ${before_gen}" + echo "super generation after balance: ${after_gen}" + echo "super generation difference: $((after_gen - before_gen))" + echo "theoretic generation difference: ${theoretic_gen}" + echo "threshold generation difference: ${threshold_gen}"
tests/btrfs/181 line 98
Quoted 10 lines
tests/btrfs/181 new file +10
-0+fi + +echo "super generation before balance: ${before_gen}" >> $seqres.full +echo "super generation after balance: ${after_gen}" >> $seqres.full +echo "super generation difference: $((after_gen - before_gen))" >> $seqres.full +echo "theoretic generation difference: ${theoretic_gen}" >> $seqres.full +echo "threshold generation difference: ${threshold_gen}" >> $seqres.full + +# success, all done +echo "Silence is golden"
Other than that, it looks good.
Thanks.
Quoted 23 lines
tests/btrfs/181 new file +3
-0+ +status=0 +exittests/btrfs/181.out new file +2
-0@@ -0,0 +1,2 @@ +QA output created by 181 +Silence is goldentests/btrfs/group +1
-1@@ -183,3 +183,4 @@ 178 auto quick send 179 auto qgroup dangerous 180 auto quick qgroup limit +181 auto quick balance --2.18.0
Signature · 3 lines
Filipe David Manana, “Whether you think you can, or you think you can't — you're right.”