| Message ID | 01796d6bcec40ae80b5af3269e60a66cd4b89262.1638382763.git.josef@toxicpanda.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fstests: btrfs/011 increase the runtime for replace cancel | expand |
On Wed, Dec 01, 2021 at 01:19:35PM -0500, Josef Bacik wrote: > This test exercises the btrfs replace cancel path, but in order to do > this we have to have enough work to do in order to successfully cancel > the balance, otherwise the test fails because the operation has > completed before we're able to cancel. This test has a very low pass > rate because we do not generate a large enough file system for replace > to have enough work, passing around 5% of the time. Increase the time > spent to 10x the time we wait for the replace to start its work before > we cancel, this allows us to consistently pass this test. I've seen the test to pass most of the time but that it still fails depending on the load is annoying and requires checking the results. So even if it's probably a workaround (not sure we can make it 100% reliable without peeking into the internals) it's making things better. Acked-by: David Sterba <dsterba@suse.com>
diff --git a/tests/btrfs/011 b/tests/btrfs/011 index b4673341..78e58a38 100755 --- a/tests/btrfs/011 +++ b/tests/btrfs/011 @@ -59,6 +59,7 @@ wait_time=1 fill_scratch() { local fssize=$1 + local with_cancel=$2 local filler_pid # Fill inline extents. @@ -87,7 +88,11 @@ fill_scratch() $XFS_IO_PROG -f -d -c "pwrite -b 64k 0 1E" "$SCRATCH_MNT/t_filler" &>\ $tmp.filler_result & filler_pid=$! - sleep $((2 * $wait_time)) + if [ "${with_cancel}" = "cancel" ]; then + sleep $((10 * $wait_time)) + else + sleep $((2 * $wait_time)) + fi kill -KILL $filler_pid &> /dev/null wait $filler_pid &> /dev/null @@ -124,7 +129,7 @@ workout() _scratch_mount _require_fs_space $SCRATCH_MNT $((2 * 512 * 1024)) #2.5G - fill_scratch $fssize + fill_scratch $fssize $with_cancel _run_btrfs_util_prog filesystem show -m $SCRATCH_MNT echo -e "Replace from $source_dev to $SPARE_DEV\\n" >> $seqres.full
This test exercises the btrfs replace cancel path, but in order to do this we have to have enough work to do in order to successfully cancel the balance, otherwise the test fails because the operation has completed before we're able to cancel. This test has a very low pass rate because we do not generate a large enough file system for replace to have enough work, passing around 5% of the time. Increase the time spent to 10x the time we wait for the replace to start its work before we cancel, this allows us to consistently pass this test. Signed-off-by: Josef Bacik <josef@toxicpanda.com> --- tests/btrfs/011 | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)