| Message ID | 20190228144128.55583-3-bfoster@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fstests: prevent dm-log-writes out of order replay issues | expand |
On Thu, Feb 28, 2019 at 4:41 PM Brian Foster <bfoster@redhat.com> wrote: > > The dm-log-writes replay mechanism issues discards to provide > zeroing functionality to prevent out-of-order replay issues. These > discards don't always result in zeroing bevavior, however, depending > on the underlying physical device. In turn, this causes test > failures on XFS v5 filesystems that enforce metadata log recovery > ordering if the filesystem ends up with stale data from the future > with respect to the active log at a particular recovery point. > > To ensure reliable discard zeroing behavior, use a thinly > provisioned volume as the data device instead of using the scratch > device directly. This slows the test down slightly, but provides > reliable functional behavior at a reduced cost from active snapshot > management or forced zeroing. > > Signed-off-by: Brian Foster <bfoster@redhat.com> Looks ok. Apart from minor nit below Reviewed-by: Amir Goldstein <amir73il@gmail.com> > --- > tests/generic/482 | 20 ++++++++++++++------ > 1 file changed, 14 insertions(+), 6 deletions(-) > > diff --git a/tests/generic/482 b/tests/generic/482 > index 3c2199d7..3b93a7fc 100755 > --- a/tests/generic/482 > +++ b/tests/generic/482 > @@ -22,12 +22,14 @@ _cleanup() > cd / > $KILLALL_PROG -KILL -q $FSSTRESS_PROG &> /dev/null > _log_writes_cleanup &> /dev/null > + _dmthin_cleanup > rm -f $tmp.* > } > > # get standard environment, filters and checks > . ./common/rc > . ./common/filter > +. ./common/dmthin > . ./common/dmlogwrites > > # remove previous $seqres.full before test > @@ -44,6 +46,7 @@ _require_command "$KILLALL_PROG" killall > _require_scratch _require_scratch_nocheck > # and we need extra device as log device > _require_log_writes > +_require_dm_target thin-pool > > > nr_cpus=$("$here/src/feature" -o) > @@ -53,9 +56,15 @@ if [ $nr_cpus -gt 8 ]; then > fi > fsstress_args=$(_scale_fsstress_args -w -d $SCRATCH_MNT -n 512 -p $nr_cpus \ > $FSSTRESS_AVOID) > +devsize=$((1024*1024*200 / 512)) # 200m phys/virt size > +csize=$((1024*64 / 512)) # 64k cluster size > +lowspace=$((1024*1024 / 512)) # 1m low space threshold > > +# Use a thin device to provide deterministic discard behavior. Discards are used > +# by the log replay tool for fast zeroing to prevent out-of-order replay issues. > _test_unmount > -_log_writes_init $SCRATCH_DEV > +_dmthin_init $devsize $devsize $csize $lowspace > +_log_writes_init $DMTHIN_VOL_DEV > _log_writes_mkfs >> $seqres.full 2>&1 > _log_writes_mark mkfs > > @@ -70,16 +79,15 @@ cur=$(_log_writes_find_next_fua $prev) > [ -z "$cur" ] && _fail "failed to locate next FUA write" > > while [ ! -z "$cur" ]; do > - _log_writes_replay_log_range $cur $SCRATCH_DEV >> $seqres.full > + _log_writes_replay_log_range $cur $DMTHIN_VOL_DEV >> $seqres.full > > # Here we need extra mount to replay the log, mainly for journal based > # fs, as their fsck will report dirty log as error. > - # We don't care to preserve any data on $SCRATCH_DEV, as we can replay > + # We don't care to preserve any data on the replay dev, as we can replay > # back to the point we need, and in fact sometimes creating/deleting > # snapshots repeatedly can be slower than replaying the log. > - _scratch_mount > - _scratch_unmount > - _check_scratch_fs > + _dmthin_mount > + _dmthin_check_fs > > prev=$cur > cur=$(_log_writes_find_next_fua $(($cur + 1))) > -- > 2.17.2 >
diff --git a/tests/generic/482 b/tests/generic/482 index 3c2199d7..3b93a7fc 100755 --- a/tests/generic/482 +++ b/tests/generic/482 @@ -22,12 +22,14 @@ _cleanup() cd / $KILLALL_PROG -KILL -q $FSSTRESS_PROG &> /dev/null _log_writes_cleanup &> /dev/null + _dmthin_cleanup rm -f $tmp.* } # get standard environment, filters and checks . ./common/rc . ./common/filter +. ./common/dmthin . ./common/dmlogwrites # remove previous $seqres.full before test @@ -44,6 +46,7 @@ _require_command "$KILLALL_PROG" killall _require_scratch # and we need extra device as log device _require_log_writes +_require_dm_target thin-pool nr_cpus=$("$here/src/feature" -o) @@ -53,9 +56,15 @@ if [ $nr_cpus -gt 8 ]; then fi fsstress_args=$(_scale_fsstress_args -w -d $SCRATCH_MNT -n 512 -p $nr_cpus \ $FSSTRESS_AVOID) +devsize=$((1024*1024*200 / 512)) # 200m phys/virt size +csize=$((1024*64 / 512)) # 64k cluster size +lowspace=$((1024*1024 / 512)) # 1m low space threshold +# Use a thin device to provide deterministic discard behavior. Discards are used +# by the log replay tool for fast zeroing to prevent out-of-order replay issues. _test_unmount -_log_writes_init $SCRATCH_DEV +_dmthin_init $devsize $devsize $csize $lowspace +_log_writes_init $DMTHIN_VOL_DEV _log_writes_mkfs >> $seqres.full 2>&1 _log_writes_mark mkfs @@ -70,16 +79,15 @@ cur=$(_log_writes_find_next_fua $prev) [ -z "$cur" ] && _fail "failed to locate next FUA write" while [ ! -z "$cur" ]; do - _log_writes_replay_log_range $cur $SCRATCH_DEV >> $seqres.full + _log_writes_replay_log_range $cur $DMTHIN_VOL_DEV >> $seqres.full # Here we need extra mount to replay the log, mainly for journal based # fs, as their fsck will report dirty log as error. - # We don't care to preserve any data on $SCRATCH_DEV, as we can replay + # We don't care to preserve any data on the replay dev, as we can replay # back to the point we need, and in fact sometimes creating/deleting # snapshots repeatedly can be slower than replaying the log. - _scratch_mount - _scratch_unmount - _check_scratch_fs + _dmthin_mount + _dmthin_check_fs prev=$cur cur=$(_log_writes_find_next_fua $(($cur + 1)))
The dm-log-writes replay mechanism issues discards to provide zeroing functionality to prevent out-of-order replay issues. These discards don't always result in zeroing bevavior, however, depending on the underlying physical device. In turn, this causes test failures on XFS v5 filesystems that enforce metadata log recovery ordering if the filesystem ends up with stale data from the future with respect to the active log at a particular recovery point. To ensure reliable discard zeroing behavior, use a thinly provisioned volume as the data device instead of using the scratch device directly. This slows the test down slightly, but provides reliable functional behavior at a reduced cost from active snapshot management or forced zeroing. Signed-off-by: Brian Foster <bfoster@redhat.com> --- tests/generic/482 | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-)