Details
-
Bug
-
Status: Reviewable
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
Mesosphere Sprint 15, Mesosphere Sprint 16, Mesosphere Sprint 17, Mesosphere Sprint 18, Mesosphere Sprint 19, Mesosphere Sprint 20, Mesosphere Sprint 21, Mesosphere Sprint 22
Description
Currently, cgroups::destroy only works properly (killing procseses + remove cgroups) if freezer subsystem is available in the hierarchy. This is rarely true. Imagine we want to remove a cgroup in memory subsystem, and if there is a process still running, the cgroups::destory will fail. The dependency on the freezer subsystem during cgroups::destroy is not strictly necessary. We can probably just loop through all the tasks in a cgroup and send SIGKILL to them until all processes are gone (by checking the cgroups procs file).
This problem caused issues when running 'sudo make check' during 0.23 release testing, where BenH provided already a better error message with b1a23d6a52c31b8c5c840ab01902dbe00cb1feef / https://reviews.apache.org/r/36604.
Attachments
Issue Links
- is related to
-
MESOS-3255 Create TasksKiller Tests
- Accepted
- relates to
-
MESOS-6414 cgroups isolator cleanup failed when the hierarchy is cleanup by docker daemon
- Resolved
-
MESOS-6489 Better support for containers that want to manage their own cgroup.
- Reviewable