On Sun, Jun 19, 2022 at 4:47 PM Amir Goldstein <amir73il@xxxxxxxxx> wrote:
>
> Those tests failed to cleanup background jobs after test
> is interrupted.
>
> Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx>
> ---
> tests/xfs/422 | 8 ++++++++
> tests/xfs/517 | 1 +
> 2 files changed, 9 insertions(+)
>
> diff --git a/tests/xfs/422 b/tests/xfs/422
> index a83a66df..8e9a3576 100755
> --- a/tests/xfs/422
> +++ b/tests/xfs/422
> @@ -13,6 +13,14 @@ _begin_fstest dangerous_scrub dangerous_online_repair freeze
>
> _register_cleanup "_cleanup" BUS
>
> +# Override the default cleanup function.
> +_cleanup()
> +{
> + $KILLALL_PROG -9 $XFS_IO_PROG $FSSTRESS_PROG > /dev/null 2>&1
> + cd /
> + rm -rf $tmp.*
> +}
> +
> # Import common functions.
> . ./common/filter
> . ./common/fuzzy
> diff --git a/tests/xfs/517 b/tests/xfs/517
> index 961668e3..18404248 100755
> --- a/tests/xfs/517
> +++ b/tests/xfs/517
> @@ -14,6 +14,7 @@ _register_cleanup "_cleanup" BUS
> # Override the default cleanup function.
> _cleanup()
> {
> + $KILLALL_PROG -9 $XFS_IO_PROG $FSSTRESS_PROG > /dev/null 2>&1
Besides the missing wait, this cleanup is still racy, because stress_loop
can spawn a new fstress process after killall /proc iteration.
Worst, freeze_loop can freeze after killall /proc iteration.
The correct solution (I think) is to record the pid of the XXX_loop
sub-shells and kill those, which should also solve the "Terminated"
false positive error.
Will give that a shot.
Thanks,
Amir.
