Ian Babrou created MESOS-2714:
---------------------------------
Summary: Memory limit for docker containers is set inconsistently
Key: MESOS-2714
URL: https://issues.apache.org/jira/browse/MESOS-2714
Project: Mesos
Issue Type: Bug
Components: docker, slave
Affects Versions: 0.22.1
Reporter: Ian Babrou
I launched 120 docker containers on unique nodes with marathon and monitoring
said that they have different memory limits.
Memory limit in marathon is set to 64mb, but 9 of 120 slaves reported limit of
96mb. Slaves are identical in terms of hardware and mesos slave versions.
I read stats from docker stats api, not from cgroup file. It turned out, that
some tasks were started with memory limit of 64mb and some with 96mb. The ones
with 64mb were increased to 96mb:
I0510 15:29:26.530024 41390 docker.cpp:1298] Updated 'cpu.shares' to 307 at
/sys/fs/cgroup/cpu/docker/b020fd33df578a9287b25886b7d9de52353fa943a6c384c4303f8bb552f377cd
for container 1e8c9f99-8519-4e35-bee6-69072f357c5e
I0510 15:29:26.530828 41390 docker.cpp:1359] Updated 'memory.limit_in_bytes' to
96MB at
/sys/fs/cgroup/memory/docker/b020fd33df578a9287b25886b7d9de52353fa943a6c384c4303f8bb552f377cd
for container 1e8c9f99-8519-4e35-bee6-69072f357c5e
In the end all tasks had 96mb limit in cgroup file, but memory limit reported
by docker was different.
I think that the limit should be set consistently and all slaves should behave
identically.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
