Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
Currently the destroyer would attempt to kill processes in a cgroup tree in parallel.
However it could result in two threads trying to freeze and thaw a parent and a child cgroup concurrently and repeatedly and it's possible that neither would make progress so it will lead to the containerizer not able to destroy the container within a timeout, after which point the agent/containerizer just gave up, leaving the (partially) frozen container holding system resources (e.g., ports).
Attachments
Issue Links
- is related to
-
MESOS-6489 Better support for containers that want to manage their own cgroup.
- Reviewable