xfs/517 fails randomally with this error: QA output created by 517 Format and populate Concurrent fsmap and freeze +Terminated Test done These two test run fsstress inside the sub-shell stress_loop(), which is run as a background job. The sub-shell has an inner loop that exist after at least 30s. The outer shell, sleeps for 32s and then kills fsstress using killall. If the inner sub-shell loop does not exit before fsstress is killed, bash sub-shell prints "Terminated" to stderr and breaks golden output. There are two easy solutions to this issue: 1. The sub-shell stderr could be redirected to /dev/null or $seq.full 2. killall can use SIGINT which suppresses the "Terminated" print The tests generic/270, generic/388, generic/475, generic/648 use the first method, but that looses any other errors that fsstress may report during the inner loop. overlay/058 uses the second method (but with SIGPIPE). Use this method to preserve other reported errors. Alas, this is not enough to fix the false positive failure - the main test thread needs to also wait for the background jobs to exit. Otherwise, killall -9 in _cleanup() will cause a similar "Killed" message in stderr. Adding -w to killall requires to move it to after unfreeze, otherwise, fsstress process may be left blocked on a frozen fs and wait will not return. Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx> --- tests/xfs/422 | 5 ++++- tests/xfs/517 | 5 ++++- 2 files changed, 8 insertions(+), 2 deletions(-) diff --git a/tests/xfs/422 b/tests/xfs/422 index 8e9a3576..63d133f8 100755 --- a/tests/xfs/422 +++ b/tests/xfs/422 @@ -95,8 +95,11 @@ while [ "$(date +%s)" -lt $((end + 2)) ]; do done # ...and clean up after the loops in case they didn't do it themselves. -$KILLALL_PROG -TERM xfs_io fsstress >> $seqres.full 2>&1 +# First thaw fs, so fsstress can exit, then kill and wait for fsstress. +# Use of SIGINT instead of SIGTERM suppresses the "Terminated" print +# from the XXX_loop() bash sub-shells $XFS_IO_PROG -x -c 'thaw' $SCRATCH_MNT >> $seqres.full 2>&1 +$KILLALL_PROG -w -SIGINT $XFS_IO_PROG $FSSTRESS_PROG >> $seqres.full 2>&1 echo "Loop finished at $(date)" >> $seqres.full echo "Test done" diff --git a/tests/xfs/517 b/tests/xfs/517 index 18404248..2f52d634 100755 --- a/tests/xfs/517 +++ b/tests/xfs/517 @@ -92,8 +92,11 @@ while [ "$(date +%s)" -lt $((end + 2)) ]; do done # ...and clean up after the loops in case they didn't do it themselves. -$KILLALL_PROG -TERM xfs_io fsstress >> $seqres.full 2>&1 +# First thaw fs, so fsstress can exit, then kill and wait for fsstress. +# Use of SIGINT instead of SIGTERM suppresses the "Terminated" print +# from the XXX_loop() bash sub-shells $XFS_IO_PROG -x -c 'thaw' $SCRATCH_MNT >> $seqres.full 2>&1 +$KILLALL_PROG -w -SIGINT $XFS_IO_PROG $FSSTRESS_PROG >> $seqres.full 2>&1 echo "Loop finished at $(date)" >> $seqres.full echo "Test done" -- 2.25.1
