Re: [PATCH tests 5/5] tests: add a regression test for raid456 deadlock
From: Mariusz Tkaczyk <hidden>
Date: 2023-05-24 09:09:58
On Tue, 23 May 2023 21:39:00 +0800 Yu Kuai [off-list ref] wrote:
quoted hunk ↗ jump to hunk
From: Yu Kuai <redacted> The deadlock is described in [1], as the last patch described, it's fixed first by [2], however this fix will be reverted and the deadlock is supposed to be fixed by [3]. [1] https://lore.kernel.org/linux-raid/5ed54ffc-ce82-bf66-4eff-390cb23bc1ac@molgen.mpg.de/T/#t (local) [2] https://lore.kernel.org/linux-raid/20220621031129.24778-1-guoqing.jiang@linux.dev/ (local) [3] https://lore.kernel.org/linux-raid/20230322064122.2384589-5-yukuai1@huaweicloud.com/ (local) Signed-off-by: Yu Kuai <redacted> --- tests/24raid456deadlock | 56 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 56 insertions(+) create mode 100644 tests/24raid456deadlockdiff --git a/tests/24raid456deadlock b/tests/24raid456deadlock new file mode 100644 index 00000000..161c3ab8 --- /dev/null +++ b/tests/24raid456deadlock@@ -0,0 +1,56 @@ +devs="$dev0 $dev1 $dev2 $dev3 $dev4 $dev5" +runtime=120 +pid="" +old=`cat /proc/sys/vm/dirty_background_ratio` + +test_write_action() +{ + while true; do + echo check > /sys/block/md0/md/sync_action &> /dev/null + sleep 0.1 + echo idle > /sys/block/md0/md/sync_action &> /dev/null + done +} + +test_write_back() +{ + fio -filename=$md0 -bs=4k -rw=write -numjobs=1 -name=test \ + -time_based -runtime=$runtime &> /dev/null +} + +set_up_test() +{ + fio -h &> /dev/null || die "fio not found" + + # create a simple raid6 + mdadm -Cv -R -n 6 -l6 $md0 $devs --assume-clean || die "create raid6failed" + + # trigger dirty pages write back + echo 0 > /proc/sys/vm/dirty_background_ratio +} + +clean_up_test() +{ + echo $old > /proc/sys/vm/dirty_background_ratio + + kill -9 $pid + sync $md0 + + if ! mdadm -S $md0; then + die "can't stop array, deadlock is probably triggered" + fi
Stop is problematic, I described why in previous patch. You can clean up array manually by ( I think you should to limit complexity): echo inactive > /sys/block/mdX/md/array_state echo clear > /sys/block/mdX/md/array_state Probably, one of those actions will hang right? The question is how we can catch it. I'm fine with current approach too: Acked-by: Mariusz Tkaczyk <redacted> Thanks, Mariusz