xfstests: test speculative preallocation reclaim on ENOSPC/EDQUOT
PATCH v3 1/1 xfstests: test speculative preallocation reclaim on ENOSPC/EDQUOT
XFS can allocate significant amounts of space to files via speculative preallocation. Such preallocation may not be reclaimed automatically on file close() if a file is repeatedly opened and extended. For smaller filesystems with relatively large and slow growing files, this preallocation can linger for some time, including contributing to out of space conditions.
Create a situation where an fs is near out of space while several files still have lingering, significant preallocations. Verify that new writers reclaim the preallocated space rather than return ENOSPC. Repeat a similar test for quota limits and EDQUOT.
---
v3:
- Assorted style fixes and cleanups.
- Removed unnecessary filtering to $seqres.full.
- Use df for free space calculation.
- Added output message.
v2: http://oss.sgi.com/archives/xfs/2014-06/msg00087.html
- Added a post-test scratch fs check.
- Use a loop device to reduce the space requirement.
- Added support for analogous EDQUOT test and added to quota group.
- Some refactoring, cleanups, etc.
v1: http://oss.sgi.com/archives/xfs/2014-05/msg00531.html
tests/xfs/014 | 200 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ tests/xfs/014.out | 2 + tests/xfs/group | 1 + 3 files changed, 203 insertions(+) create mode 100755 tests/xfs/014 create mode 100644 tests/xfs/014.out
tests/xfs/014 new file +200 -0
@@ -0,0 +1,200 @@ +#!/bin/bash +# FS QA Test No. xfs/014 +# +# Test the behavior of XFS dynamic speculative preallocation at ENOSPC and +# EDQUOT conditions. Speculative preallocation allocates post-EOF space to files +# as they are extended. This test creates conditions where an fs is near a space +# limit with lingering, relatively significant preallocations and verifies that +# new writers reclaim said preallocations rather than prematurely fail with +# ENOSPC/EDQUOT. +# +#----------------------------------------------------------------------- +# Copyright (c) 2014 Red Hat, Inc. All Rights Reserved. +# +# This program is free software; you can redistribute it and/or +# modify it under the terms of the GNU General Public License as +# published by the Free Software Foundation. +# +# This program is distributed in the hope that it would be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program; if not, write the Free Software Foundation, +# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA +# +#----------------------------------------------------------------------- +# + +seq=`basename $0` +seqres=$RESULT_DIR/$seq +echo "QA output created by $seq" + +here=`pwd` +tmp=/tmp/$$ +status=1 # failure is the default! + +# get standard environment, filters and checks +. ./common/rc +. ./common/filter +. ./common/quota + +_cleanup() +{ + cd / + umount $LOOP_MNT 2>/dev/null + umount $SCRATCH_MNT 2>/dev/null + rm -f $tmp.* +} +trap "_cleanup; exit \$status" 0 1 2 3 15 + +# Create a file using a repeated open, extending write and close pattern. This +# causes the preallocation to persist after the file is closed. Preallocation +# will not be reclaimed unless the inode is evicted or we hit an allocation +# failure. +_spec_prealloc_file() +{ + file=$1 + + rm -f $file + + # a few file extending open-write-close cycles should be enough to + # trigger the fs to retain preallocation. write 256k in 32k intervals to + # be sure + for i in $(seq 0 32768 262144); do + $XFS_IO_PROG -f -c "pwrite $i 32k" $file >> $seqres.full + done + + # write a 4k aligned amount of data to keep the calculations simple + $XFS_IO_PROG -c "pwrite 0 128m" $file >> $seqres.full + + size=`stat -c "%s" $file` + blocks=`stat -c "%b" $file` + blocksize=`stat -c "%B" $file` + + prealloc_size=$((blocks * blocksize - size)) + if [ $prealloc_size -eq 0 ]; then + echo "Warning: No speculative preallocation for $file." \ + "Check use of the allocsize= mount option." + fi + + # keep a running total of how much preallocation we've created + TOTAL_PREALLOC=$((TOTAL_PREALLOC + prealloc_size)) +} + +_consume_free_space() +{ + dir=$1 + + # allocate all but 10MB of available space + freesp=`df -m $dir | awk '/^\// { print $4 - 10 }'` + $XFS_IO_PROG -f -c "falloc 0 ${freesp}M" $dir/spc +} + +# Create several files with preallocation and consume the remaining free space +# via fallocate to the put the fs at ENOSPC. Create a set of background writers +# to write into ENOSPC and cause the preallocation to be reclaimed and +# reallocated to the new writers. +_test_enospc() +{ + dir=$1 + + rm -rf $dir/* + + TOTAL_PREALLOC=0 + for i in $(seq 0 3); do + _spec_prealloc_file $dir/pre$i + done + + _consume_free_space $dir + + # consume 1/2 of the current preallocation across the set of 4 writers + write_size=$((TOTAL_PREALLOC / 2 / 4)) + for i in $(seq 0 3); do + $XFS_IO_PROG -f -c "pwrite 0 $write_size" $dir/file.$i \ + >> $seqres.full & + done + + wait +} + +# Create preallocations accounted by both user and group quotas. Set the +# associated quota hard limits to put them at EDQUOT. Verify that a new writer +# reclaims the preallocated space and proceeds without error. +_test_edquot() +{ + dir=$1 + + rm -rf $dir/* + + TOTAL_PREALLOC=0 + _spec_prealloc_file $dir/user + chown $qa_user $dir/user + + _spec_prealloc_file $dir/group + chgrp $qa_group $dir/group + + # writing to a file under both quotas means both will be reclaimed on + # allocation failure + touch $dir/file + chown $qa_user $dir/file + chgrp $qa_group $dir/file + + # put both quotas at EDQUOT + blks=`$XFS_QUOTA_PROG -xc "quota -u $qa_user" $dir | \ + tail -n 1 | awk '{ print $2 }'` + $XFS_QUOTA_PROG -xc "limit -u bhard=${blks}k $qa_user" $dir + blks=`$XFS_QUOTA_PROG -xc "quota -g $qa_grup" $dir | \ + tail -n 1 | awk '{ print $2 }'` + $XFS_QUOTA_PROG -xc "limit -g bhard=${blks}k $qa_group" $dir + + # each quota has a single file worth of preallocation to reclaim. leave + # some wiggle room and write to 1/3 the total. + write_size=$((TOTAL_PREALLOC / 3)) + $XFS_IO_PROG -c "pwrite 0 $write_size" $dir/file >> $seqres.full +} + +# real QA test starts here +_supported_fs xfs +_supported_os Linux + +_require_scratch +_require_xfs_io_command "falloc" +_require_loop +_require_quota +_require_user +_require_group + +rm -f $seqres.full + +echo "Silence is golden." + +_scratch_mkfs_xfs >> $seqres.full 2>&1 +_scratch_mount + +# make sure the background eofblocks scanner doesn't interfere +orig_sp_time=`cat /proc/sys/fs/xfs/speculative_prealloc_lifetime` +echo 9999 > /proc/sys/fs/xfs/speculative_prealloc_lifetime + +LOOP_FILE=$SCRATCH_MNT/$seq.fs +LOOP_MNT=$SCRATCH_MNT/$seq.mnt + +$MKFS_XFS_PROG -d "file=1,name=$LOOP_FILE,size=10g" >> $seqres.full 2>&1 + +mkdir -p $LOOP_MNT +mount -t xfs -o loop,uquota,gquota $LOOP_FILE $LOOP_MNT || \ + _fail "Failed to mount loop fs." + +_test_enospc $LOOP_MNT +_test_edquot $LOOP_MNT + +umount $LOOP_MNT + +echo $orig_sp_time > /proc/sys/fs/xfs/speculative_prealloc_lifetime + +umount $SCRATCH_MNT +_check_scratch_fs + +status=0 +exit
tests/xfs/014.out new file +2 -0
@@ -0,0 +1,2 @@ +QA output created by 014 +Silence is golden.
tests/xfs/group +1 -0
@@ -10,6 +10,7 @@ 010 auto quick repair 012 rw auto quick 013 auto metadata stress +014 auto enospc quick quota 016 rw auto quick 017 mount auto quick stress 018 deprecated # log logprint v2log
Signature · 1 line
1.8.3.1
Quoted 7 lines · Brian Foster
XFS can allocate significant amounts of space to files via speculative preallocation. Such preallocation may not be reclaimed automatically on file
close()if a file is repeatedly opened and extended. For smaller filesystems with relatively large and slow growing files, this preallocation can linger for some time, including contributing to out of space conditions.
Quoted 6 lines
Create a situation where an fs is near out of space while several files still have lingering, significant preallocations. Verify that new writers reclaim the preallocated space rather than return ENOSPC. Repeat a similar test for quota limits and EDQUOT.
Signed-off-by Brian Foster <[email protected]>
Hi Brian,
My test machines all fail this test with output like this:
xfs/014 - output mismatch (see /home/dave/src/xfstests-dev/results//xfs/014.out.bad)
--- tests/xfs/014.out 2014-06-18 09:32:59.000000000 +1000
+++ /home/dave/src/xfstests-dev/results//xfs/014.out.bad 2014-06-18 10:27:10.000000000 +1000
@@ -1,2 +1,7 @@
QA output created by 014
Silence is golden.
+/mnt/scratch/014.mnt/file.0: No space left on device
+pwrite64: No space left on device
+/mnt/scratch/014.mnt/file.2: No space left on device
+pwrite64: No space left on device
+pwrite64: Disk quota exceeded
.....
or this from a 1k block size filesystem:
xfs/014 - output mismatch (see /home/dave/src/xfstests-dev/results//xfs/014.out.bad)
--- tests/xfs/014.out 2014-06-18 09:32:59.000000000 +1000
+++ /home/dave/src/xfstests-dev/results//xfs/014.out.bad 2014-06-18 10:29:15.000000000 +1000
@@ -1,2 +1,7 @@
QA output created by 014
Silence is golden.
+pwrite64: No space left on device
+pwrite64: No space left on device
+pwrite64: No space left on device
+pwrite64: No space left on device
+pwrite64: Disk quota exceeded
.....
I'm still going to commit the test as it stands, but could you see if you can reproduce this or suggest hints as to where it might be going wrong?
FWIW, patches to tee the stderr output to both the golden output and the $seqres.full file will make it much easier to determine what write is failing....
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
Quoted 13 lines · Dave Chinner
Quoted 12 lines · Brian Foster
XFS can allocate significant amounts of space to files via speculative preallocation. Such preallocation may not be reclaimed automatically on file
close()if a file is repeatedly opened and extended. For smaller filesystems with relatively large and slow growing files, this preallocation can linger for some time, including contributing to out of space conditions.Create a situation where an fs is near out of space while several files still have lingering, significant preallocations. Verify that new writers reclaim the preallocated space rather than return ENOSPC. Repeat a similar test for quota limits and EDQUOT.
Quoted 6 lines
Quoted 1 line
Signed-off-by Brian Foster <[email protected]>Hi Brian,
My test machines all fail this test with output like this:
This is most likely due to not having the corresponding eofblocks scan on enospc patches (as noted on irc). The test creates the conditions that currently lead to ENOSPC and tests that the scan frees up enough space to allow writes to proceed.
Let me know if you continue to reproduce this and I'll dig further into it...
Brian
Quoted 42 lines
xfs/014 - output mismatch (see
/home/dave/src/xfstests-dev/results//xfs/014.out.bad)
--- tests/xfs/014.out 2014-06-18 09:32:59.000000000 +1000
+++/home/dave/src/xfstests-dev/results//xfs/014.out.bad2014-06-18 10:27:10.000000000 +1000
@@ -1,2 +1,7 @@
QA output created by 014
Silence is golden.
+/mnt/scratch/014.mnt/file.0: No space left on device
+pwrite64: No space left on device
+/mnt/scratch/014.mnt/file.2: No space left on device
+pwrite64: No space left on device
+pwrite64: Disk quota exceeded
.....or this from a 1k block size filesystem:
xfs/014 - output mismatch (see
/home/dave/src/xfstests-dev/results//xfs/014.out.bad)
--- tests/xfs/014.out 2014-06-18 09:32:59.000000000 +1000
+++/home/dave/src/xfstests-dev/results//xfs/014.out.bad2014-06-18 10:29:15.000000000 +1000
@@ -1,2 +1,7 @@
QA output created by 014
Silence is golden.
+pwrite64: No space left on device
+pwrite64: No space left on device
+pwrite64: No space left on device
+pwrite64: No space left on device
+pwrite64: Disk quota exceeded
.....I'm still going to commit the test as it stands, but could you see if you can reproduce this or suggest hints as to where it might be going wrong?
FWIW, patches to tee the stderr output to both the golden output and the $seqres.full file will make it much easier to determine what write is failing....
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]