Conversation
linux/task.go
Outdated
There was a problem hiding this comment.
Can we not IgnoreNotExist error here, and return ErrNotFound error?
@crosbymichael
There was a problem hiding this comment.
This is hard because it will error if one subsystem does not exit. We would have to modify the cgroup code more to see if none of the cgroups exits.
There was a problem hiding this comment.
I see. We've handled empty stats in cri-containerd containerd/cri#328, so it's fine. :)
Codecov Report
@@ Coverage Diff @@
## master #1598 +/- ##
=======================================
Coverage 46.29% 46.29%
=======================================
Files 24 24
Lines 3378 3378
=======================================
Hits 1564 1564
Misses 1456 1456
Partials 358 358Continue to review full report at Codecov.
|
|
@Random-Liu do you want to include the cgroups package bump in this or do you want me to update that in a separate PR? |
|
@crosbymichael I'll do soon. |
Signed-off-by: Lantao Liu <[email protected]>
Random-Liu
force-pushed
the
fix-load-task
branch
from
989cf4c to
28ca8f0
Compare
|
@crosbymichael Done. |
|
LGTM |
Fix a Rootless Docker-in-Docker issue on Fedora 30: docker-library/docker#165 (comment) Related: containerd#1598 Signed-off-by: Akihiro Suda <[email protected]>
Fix a Rootless Docker-in-Docker issue on Fedora 30: docker-library/docker#165 (comment) Related: containerd#1598 Signed-off-by: Akihiro Suda <[email protected]> (cherry picked from commit fab016c) Signed-off-by: Akihiro Suda <[email protected]>
Fix a Rootless Docker-in-Docker issue on Fedora 30: docker-library/docker#165 (comment) Related: containerd#1598 Signed-off-by: Akihiro Suda <[email protected]>
4 participants
Rely on containerd/cgroups#30.
Currently, after runc container exits, the cgroups will be gone, and
cgroups.Loadwill always return error.This means that we'll never be able to reload a stopped
Task:This PR ignores
cgroups.ErrCgroupDeletedand fixes other places correspondingly.Actually, it would be ideal if
Metricscould always returnErrNotFoundafter runc container dies, but that is not the case today even with this PR.Signed-off-by: Lantao Liu [email protected]