On 2021/03/25 11:15, Shin'ichiro Kawasaki wrote:
> A recently fixed bug was caused by zone reset during asynchronous IOs
> in-flight. The bug symptom was unaligned command error which was
> observed using random write workload with libaio engine and block size
> not a divisor of zone size. To confirm the bug fix and to prevent future
> regression, add a test case which runs the workload.
>
> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@xxxxxxx>
> ---
> t/zbd/test-zbd-support | 18 ++++++++++++++++++
> 1 file changed, 18 insertions(+)
>
> diff --git a/t/zbd/test-zbd-support b/t/zbd/test-zbd-support
> index be129615..51c92e86 100755
> --- a/t/zbd/test-zbd-support
> +++ b/t/zbd/test-zbd-support
> @@ -1201,6 +1201,24 @@ test56() {
> >> "${logfile}.${test_number}" 2>&1 || return $?
> }
>
> +# Test that repeated async write job does not cause zone reset during writes
> +# in-flight, when the block size is not a divisor of the zone size.
> +test57() {
> + local bs off
> +
> + require_zbd || return $SKIP_TESTCASE
> +
> + bs=$((4096 * 7))
> + off=$((first_sequential_zone_sector * 512))
> +
> + run_fio --name=job --filename="${dev}" --rw=randwrite --bs="${bs}" \
> + --offset="${off}" --size=$((4 * zone_size)) --iodepth=256 \
> + "$(ioengine "libaio")" --numjob=4 --time_based=1 --runtime=30s \
With a single job, there will be no lock contention, so more chances of seeing
the problem as you described in the cover letter. So may be remove numjobs=4 here ?
> + --zonemode=zbd --direct=1 --zonesize="${zone_size}" \
> + ${job_var_opts[@]} \
> + >> "${logfile}.${test_number}" 2>&1 || return $?
> +}
> +
> SECONDS=0
> tests=()
> dynamic_analyzer=()
>
--
Damien Le Moal
Western Digital Research
