Skip to content

cgroup related tests seem to be failing more often in CI #1774

@adrianreber

Description

@adrianreber
2022-03-11T14:49:33.7079636Z b'(00.022093) cg: adding cgroup /proc/self/fd/15/docker/41af214ff35599b069c697d142ceea1813ac16aceb1cc6b389cdc72d72120cce'
2022-03-11T14:49:33.7080078Z b'(00.022301) cg: adding cgroup /proc/self/fd/15/'
2022-03-11T14:49:33.7080558Z b'(00.022429) cg: adding cgroup /proc/self/fd/15/docker/41af214ff35599b069c697d142ceea1813ac16aceb1cc6b389cdc72d72120cce'
2022-03-11T14:49:33.7081009Z b'(00.022557) cg: adding cgroup /proc/self/fd/15/subcg00/subsubcg'
2022-03-11T14:49:33.7081501Z b'(00.040339) Error (criu/cgroup.c:587): cg: Unable to create the cgroup (name=zdtmtst.defaultroot) file system: Resource busy'
2022-03-11T14:49:33.7082051Z b'(00.040374) Error (criu/cgroup.c:715): cg: failed walking /proc/self/fd/-1/ for empty cgroups: No such file or directory'

Recently I have seen more test failures concerning cgroup in CI. CI seems to have changed to a 5.13 although I am not sure that is related.

The error is easy to reproduce and I can see it locally about 10% of the time.

It seems to not happen if I disabled the new mount API support.

I see that ret = sys_fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0); sometimes returns EBUSY. I tried to wrap it in a loop to retry if it fails with EBUSY, but the second time it then fails with EINVAL. For some reason the kernel behaves differently if I repeat the following steps: fsopen(), fsconfig(), fsconfig(), close().

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions