On Mon, May 25, 2020 at 12:46:48AM +0800, Eryu Guan wrote:
> On Wed, May 20, 2020 at 01:52:58PM +1000, Donald Douwsma wrote:
> > xfs_repair's interval based progress has been broken for
> > some time, create a test based on dmdelay to stretch out
> > the time and use ag_stride to force parallelism.
> >
> > Signed-off-by: Donald Douwsma <ddouwsma@xxxxxxxxxx>
> > ---
> > Changes since v1:
> > - Use _scratch_xfs_repair
> > - Filter only repair output
> > - Make the filter more tolerant of whitespace and plurals
> > - Take golden output from 'xfs_repair: fix progress reporting'
>
> I saw failures like below, and I'm using v5.7-rc4 kernel and v5.4.0
> xfsprogs, is this expected failure?
>
> @@ -2,8 +2,6 @@
> Format and populate
> Introduce a dmdelay
> Run repair
> - - #:#:#: Phase #: #% done - estimated remaining time # minute, # second
> - - #:#:#: Phase #: elapsed time # second - processed # inodes per minute
> - #:#:#: check for inodes claiming duplicate blocks - # of # inodes done
> - #:#:#: process known inodes and inode discovery - # of # inodes done
> - #:#:#: process newly discovered inodes - # of # allocation groups done
> @@ -12,4 +10,3 @@
> - #:#:#: scanning filesystem freespace - # of # allocation groups done
> - #:#:#: setting up duplicate extent list - # of # allocation groups done
> - #:#:#: verify and correct link counts - # of # allocation groups done
> - - #:#:#: zeroing log - # of # blocks done
Yeah, there's an xfsprogs patch pending, IIRC.
--D
>
> >
> > tests/xfs/516 | 76 +++++++++++++++++++++++++++++++++++++++++++++++
> > tests/xfs/516.out | 15 ++++++++++
> > tests/xfs/group | 1 +
> > 3 files changed, 92 insertions(+)
> > create mode 100755 tests/xfs/516
> > create mode 100644 tests/xfs/516.out
> >
> > diff --git a/tests/xfs/516 b/tests/xfs/516
> > new file mode 100755
> > index 00000000..1c0508ef
> > --- /dev/null
> > +++ b/tests/xfs/516
> > @@ -0,0 +1,76 @@
> > +#! /bin/bash
> > +# SPDX-License-Identifier: GPL-2.0
> > +# Copyright (c) 2020 Red Hat, Inc. All Rights Reserved.
> > +#
> > +# FS QA Test 516
> > +#
> > +# Test xfs_repair's progress reporting
> > +#
> > +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()
> > +{
>
> rm -f $tmp.*
>
> As some common helpers would use $tmp. files as well.
>
> > + cd /
> > + _dmsetup_remove delay-test > /dev/null 2>&1
>
> I think we could do _cleanup_delay here and discard the outputs.
>
> > +}
> > +
> > +# get standard environment, filters and checks
> > +. ./common/rc
> > +. ./common/filter
> > +. ./common/dmdelay
> > +. ./common/populate
> > +
> > +# remove previous $seqres.full before test
> > +rm -f $seqres.full
> > +
> > +# real QA test starts here
> > +
> > +# Modify as appropriate.
> > +_supported_fs xfs
> > +_supported_os Linux
> > +_require_scratch
> > +_require_dm_target delay
> > +
> > +# Filter output specific to the formatters in xfs_repair/progress.c
> > +# Ideally we'd like to see hits on anything that matches
> > +# awk '/{FMT/' repair/progress.c
> > +_filter_repair()
>
> Function names with the leading underscore are reserved for common
> helpers, filter_repair would be fine.
>
> > +{
> > + sed -ne '
> > + s/[0-9]\+/#/g;
> > + s/^\s\+/ /g;
> > + s/\(second\|minute\)s/\1/g
> > + /#:#:#:/p
> > + '
> > +}
> > +
> > +echo "Format and populate"
> > +_scratch_populate_cached nofill > $seqres.full 2>&1
> > +
> > +echo "Introduce a dmdelay"
> > +_init_delay
> > +
> > +# Introduce a read I/O delay
> > +# The default in common/dmdelay is a bit too agressive
> > +BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
> > +DELAY_TABLE_RDELAY="0 $BLK_DEV_SIZE delay $SCRATCH_DEV 0 100 $SCRATCH_DEV 0 0"
> > +_load_delay_table $DELAY_READ
> > +
> > +echo "Run repair"
> > +SCRATCH_DEV=$DELAY_DEV _scratch_xfs_repair -o ag_stride=4 -t 1 2>&1 |
> > + tee -a $seqres.full > $seqres.xfs_repair.out
> > +
> > +cat $seqres.xfs_repair.out | _filter_repair | sort -u
>
> I agreed with Darrick here. redirect output to $tmp.repair is better, as
> we already cleanup $tmp.* in _cleanup, and no one is cleaning up
> $seqres.xfs_repair.out file.
>
> > +
> > +_cleanup_delay
>
> We could remove this one if do it in _cleanup.
>
> > +
> > +# success, all done
> > +status=0
> > +exit
> > diff --git a/tests/xfs/516.out b/tests/xfs/516.out
> > new file mode 100644
> > index 00000000..85018b93
> > --- /dev/null
> > +++ b/tests/xfs/516.out
> > @@ -0,0 +1,15 @@
> > +QA output created by 516
> > +Format and populate
> > +Introduce a dmdelay
> > +Run repair
> > + - #:#:#: Phase #: #% done - estimated remaining time # minute, # second
> > + - #:#:#: Phase #: elapsed time # second - processed # inodes per minute
> > + - #:#:#: check for inodes claiming duplicate blocks - # of # inodes done
> > + - #:#:#: process known inodes and inode discovery - # of # inodes done
> > + - #:#:#: process newly discovered inodes - # of # allocation groups done
> > + - #:#:#: rebuild AG headers and trees - # of # allocation groups done
> > + - #:#:#: scanning agi unlinked lists - # of # allocation groups done
> > + - #:#:#: scanning filesystem freespace - # of # allocation groups done
> > + - #:#:#: setting up duplicate extent list - # of # allocation groups done
> > + - #:#:#: verify and correct link counts - # of # allocation groups done
> > + - #:#:#: zeroing log - # of # blocks done
> > diff --git a/tests/xfs/group b/tests/xfs/group
> > index 12eb55c9..aeeca23f 100644
> > --- a/tests/xfs/group
> > +++ b/tests/xfs/group
> > @@ -513,3 +513,4 @@
> > 513 auto mount
> > 514 auto quick db
> > 515 auto quick quota
> > +516 repair
>
> Should be in auto group as well? Only tests in auto (and quick, which is
> a sub-set of auto) will be run by default.
>
> Thanks,
> Eryu
>
> > --
> > 2.18.4
