Re: [RFC PATCH v2 3/3] xfs: stress test for shrinking free space in the last AG

[Gao Xiang <hsiangkao@redhat.com>]

On Fri, Mar 12, 2021 at 08:37:13AM -0800, Darrick J. Wong wrote:
> On Sat, Mar 13, 2021 at 12:17:44AM +0800, Gao Xiang wrote:
> > On Sat, Mar 13, 2021 at 12:17:55AM +0800, Zorro Lang wrote:
> > > On Fri, Mar 12, 2021 at 09:23:00PM +0800, Gao Xiang wrote:
> > > > This adds a stress testcase to shrink free space as much as
> > > > possible in the last AG with background fsstress workload.
> > > > 
> > > > The expectation is that no crash happens with expected output.
> > > > 
> > > > Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
> > > > ---
> > > > Note that I don't use _fill_fs instead, since fill_scratch here mainly to
> > > > eat 125M to make fsstress more effectively, rather than fill data as
> > > > much as possible.
> > > 
> > > As Darrick had given lots of review points to this case, I just have
> > > 2 picky questions as below:)
> > > 
> > > > 
> > > >  tests/xfs/991     | 121 ++++++++++++++++++++++++++++++++++++++++++++++
> > > >  tests/xfs/991.out |   8 +++
> > > >  tests/xfs/group   |   1 +
> > > >  3 files changed, 130 insertions(+)
> > > >  create mode 100755 tests/xfs/991
> > > >  create mode 100644 tests/xfs/991.out
> > > > 
> > > > diff --git a/tests/xfs/991 b/tests/xfs/991
> > > > new file mode 100755
> > > > index 00000000..22a5ac81
> > > > --- /dev/null
> > > > +++ b/tests/xfs/991
> > > > @@ -0,0 +1,121 @@
> > > > +#! /bin/bash
> > > > +# SPDX-License-Identifier: GPL-2.0
> > > > +# Copyright (c) 2020-2021 Red Hat, Inc.  All Rights Reserved.
> > > > +#
> > > > +# FS QA Test 991
> > > > +#
> > > > +# XFS online shrinkfs stress test
> > > > +#
> > > > +# This test attempts to shrink unused space as much as possible with
> > > > +# background fsstress workload. It will decrease the shrink size if
> > > > +# larger size fails. And totally repeat 2 * TIME_FACTOR times.
> > > > +#
> > > > +seq=`basename $0`
> > > > +seqres=$RESULT_DIR/$seq
> > > > +echo "QA output created by $seq"
> > > > +
> > > > +here=`pwd`
> > > > +tmp=/tmp/$$
> > > > +status=1	# failure is the default!
> > > > +trap "rm -f $tmp.*; exit \$status" 0 1 2 3 15
> > > > +
> > > > +# get standard environment, filters and checks
> > > > +. ./common/rc
> > > > +. ./common/filter
> > > > +
> > > > +create_scratch()
> > > > +{
> > > > +	_scratch_mkfs_xfs $@ | tee -a $seqres.full | \
> > > > +		_filter_mkfs 2>$tmp.mkfs >/dev/null
> > > > +	. $tmp.mkfs
> > > > +
> > > > +	if ! _try_scratch_mount 2>/dev/null; then
> > > > +		echo "failed to mount $SCRATCH_DEV"
> > > > +		exit 1
> > > > +	fi
> > > > +
> > > > +	# fix the reserve block pool to a known size so that the enospc
> > > > +	# calculations work out correctly.
> > > > +	_scratch_resvblks 1024 > /dev/null 2>&1
> > > > +}
> > > > +
> > > > +fill_scratch()
> > > > +{
> > > > +	$XFS_IO_PROG -f -c "resvsp 0 ${1}" $SCRATCH_MNT/resvfile
> > > > +}
> > > > +
> > > > +stress_scratch()
> > > > +{
> > > > +	procs=3
> > > > +	nops=$((1000 * LOAD_FACTOR))
> > > > +	# -w ensures that the only ops are ones which cause write I/O
> > > > +	FSSTRESS_ARGS=`_scale_fsstress_args -d $SCRATCH_MNT -w -p $procs \
> > > > +	    -n $nops $FSSTRESS_AVOID`
> > > > +	$FSSTRESS_PROG $FSSTRESS_ARGS >> $seqres.full 2>&1 &
> > > > +}
> > > > +
> > > > +# real QA test starts here
> > > > +_supported_fs xfs
> > > > +_require_scratch
> > > > +_require_xfs_shrink
> > > > +_require_xfs_io_command "falloc"
> > > 
> > > Do I miss something? I only found you use xfs_io "resvsp", why you need "falloc" cmd?
> > 
> > As I mentioned before, the testcase was derived from xfs/104 with some
> > modification.
> > 
> > At a quick glance, this line was added by commit 09e94f84d929 ("xfs: don't
> > assume preallocation is always supported on XFS"). I have no more background
> > yet.
> 
> Why not use xfs_io falloc in the test?  fallocate is the successor to
> resvsp.

Yeah, general falloc seems better, and it seems _require_xfs_io_command here is
used for always_cow inode feature. Will update it. Thanks!

Thanks,
Gao Xiang

> 
> --D