Re: [PATCH v4 01/15] btrfs: new test to run btrfs balance and subvolume test simultaneously

[Eryu Guan <eguan@redhat.com>]

On Mon, Sep 29, 2014 at 03:03:48PM -0400, Josef Bacik wrote:
> On 09/26/2014 12:14 AM, Eryu Guan wrote:
> >Run btrfs balance and subvolume create/mount/umount/delete simultaneously,
> >with fsstress running in background.
> >
> >Signed-off-by: Eryu Guan <eguan@redhat.com>
> >---
[snip]
> >+	args=`_scale_fsstress_args -p 20 -n 100 $FSSTRESS_AVOID -d $SCRATCH_MNT/stressdir`
> >+	echo "Run fsstress $args" >>$seqres.full
> >+	$FSSTRESS_PROG $args >/dev/null 2>&1 &
> >+	fsstress_pid=$!
> >+
> >+	echo -n "Start balance worker: " >>$seqres.full
> >+	_btrfs_stress_balance $SCRATCH_MNT >/dev/null 2>&1 &
> >+	balance_pid=$!
> >+	echo "$balance_pid" >>$seqres.full
> >+
> >+	echo -n "Start subvolume worker: " >>$seqres.full
> >+	_btrfs_stress_subvolume $SCRATCH_DEV $SCRATCH_MNT subvol_$$ $subvol_mnt >/dev/null 2>&1 &
> >+	subvol_pid=$!
> >+	echo "$subvol_pid" >>$seqres.full
> >+
> >+	echo "Wait for fsstress to exit and kill all background workers" >>$seqres.full
> >+	wait $fsstress_pid
> >+
> >+	kill $balance_pid $subvol_pid
> >+	wait
> >+	# wait for the balance operation to finish
> >+	while ps aux | grep "balance start" | grep -qv grep; do
> >+		sleep 1
> >+	done
> 
> This bit isn't needed, killing the balance pid also won't do anything, just
> waiting is enough, it'll only exit once the balance is finished. If you
> really want to stop the balance you can use

I think it's still needed, or the balance process would block the
later umount and test fails like

--- tests/btrfs/059.out 2014-09-26 11:35:43.932850826 +0800
+++ /root/xfstests/results//btrfs/059.out.bad   2014-09-30 23:29:47.538310375 +0800
@@ -1,2 +1,9 @@
 QA output created by 059
 Silence is golden
+umount: /mnt/testarea/scratch: target is busy.
+        (In some cases useful info about processes that use
+         the device is found by lsof(8) or fuser(1))
+umount: /mnt/testarea/scratch: target is busy.
+        (In some cases useful info about processes that use
+         the device is found by lsof(8) or fuser(1))
+_check_btrfs_filesystem: filesystem on /dev/mapper/rhel_hp--dl388eg8--01-testlv2 is inconsistent (see /root/xfstests/results//btrfs/059.full)

Adding debug output of "lsof $SCRATCH_MNT" and "ps aux | grep btrfs"
before umount shows it's the balance process is still running and
blocks the umount.

lsof /mnt/testarea/scratch
btrfs   13491 root    3r   DIR   0,40       42  256 /mnt/testarea/scratch
ps aux | grep btrfs
....
root     13491  7.0  0.0  18660  1312 pts/1    D+   23:29   0:00 /usr/sbin/btrfs balance start /mnt/testarea/scratch

Killing the balance pid is to stop any further balance operation, wait
is waiting for the child process (_btrfs_stress_balance), not the
child's child process (balance process). So only wait is not enough,
we should wait for the balance to finish explicitly.

This is also true for the rest of the test cases in this series. This
is kind of ugly, but I cannot figure out a better solution right now..

Thanks,
Eryu