On Tue, Mar 04, 2025 at 10:46:40AM +0800, Yu Kuai wrote: > Hi, > > 在 2025/02/25 10:07, Yu Kuai 写道: > > Hi, > > > > 在 2025/02/24 21:54, Ming Lei 写道: > > > > Do you run this test without the kernel patch? If I remembered > > > > correctly, ext4 issue the meta IO from jbd2 by default, which is from > > > > root cgroup, and root cgroup can only throttled from cgroup v1. > > > It passes on v6.14-rc1, does META/SWAP IO only route from cgroup v1? > > > > Of course not, it's just ext4 will issue such IO from root cgroup, and > > there is no IO can be throttled here. > > > > You might want to bind jbd2 to the test cgroup as well. But the issue still can't be reproduced by adding the following delta change, meantime revert 29390bb5661d ("blk-throttle: support prioritized processing of metadata") on kernel side. diff --git a/tests/throtl/006 b/tests/throtl/006 index 4baadaf..bb09eb2 100755 --- a/tests/throtl/006 +++ b/tests/throtl/006 @@ -43,7 +43,11 @@ test() { _throtl_set_limits wbps=$((1024 * 1024)) { + local jbd2_pid + + jbd2_pid=$(ps -eo pid,comm |grep "jbd2/${THROTL_DEV}" |awk '{print $1}') echo "$BASHPID" > "$CGROUP2_DIR/$THROTL_DIR/cgroup.procs" + echo "$jbd2_pid" > "$CGROUP2_DIR/$THROTL_DIR/cgroup.procs" _throtl_issue_fs_io "${TMPDIR}/mnt/test.img" write 64K 64 & sleep 2 test_meta_io "${TMPDIR}/mnt" Thanks, Ming