[MESOS-3488] /sys/fs/cgroup/memory/mesos missing after running a while

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Invalid
• Affects Version/s: 0.21.0
• Labels: systemd

Environment

mesos 0.21.0 on CentOS 7.1

• Epic Link: systemd
• Assignee: Unassigned
• Reporter: Chengwei Yang
• Created: 22/Sep/15 06:43
• Updated: 26/Sep/17 18:41
• Resolved: 22/Sep/15 10:45

Description

I setup mesos 0.21.0 on CentOS 7.1 with mesos-0.21.0 rpm downloaded from mesosphere.

at first, it works fine, jobs are finished correctly, however, after running a while, all task goes to LOST.

And there is nothing in *sandbox* and I see below from mesos-slave.ERROR

```
E0922 14:02:31.329264 8336 slave.cpp:2787] Container '865c9a1c-3abe-4263-921e-8d0be2f0a56d' for executor '21bf1400-6456-45e5-8f28-fe5ade7e7bfd' of framework '20150401-105258-3755085578-5050-14676-0000' failed to start: Failed to prepare isolator: Failed to create directory '/sys/fs/cgroup/memory/mesos/865c9a1c-3abe-4263-921e-8d0be2f0a56d': No such file or directory
```

And I checked that /sys/fs/cgroup/cpu/mesos does exist while /sys/fs/cgroup/memory/mesos is missing.

Since at first mesos works fine (and I checked source code, that if it create /sys/fs/cgroup/memory/mesos failed at mesos-slave startup, it will log that error), so I'm curious when/which removed /sys/fs/cgroup/memory/mesos.

Has anyone saw this issue before?

Issue Links

• is related to: MESOS-7990 Support systemd named hierarchy (name=systemd) for Mesos Containerizer.
• is related to: MESOS-3009 Reproduce systemd cgroup behavior