[
https://issues.apache.org/jira/browse/MESOS-3333?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15435937#comment-15435937
]
Christopher Hunt commented on MESOS-3333:
-----------------------------------------
Given that an executor and any of its tasks share the same cgroup limits, and
the OOM killer kills processes that exceed those limits, I think that it is
essential that an executor's tasks are able to be managed in their own cgroups.
Our use case is that we have a custom executor responsible for many tasks. If
one of those tasks is not correctly dimensioned in terms of its resource
requirements then I understand that all tasks being managed by that executor
will be brought down. Given that we do not control the dimensioning of these
resource requirements, our framework's users do, then this provides an
unsatisfactory experience when using our framework.
> Allow executors to create nested containers.
> --------------------------------------------
>
> Key: MESOS-3333
> URL: https://issues.apache.org/jira/browse/MESOS-3333
> Project: Mesos
> Issue Type: Epic
> Components: containerization, slave
> Reporter: Connor Doyle
> Labels: mesosphere
>
> Allow executors to create and manage nested containers. These nested
> containers should be isolated and monitored in the same manner as executor
> containers are currently managed.
> Some high-level thoughts and ideas are captured here:
> https://docs.google.com/document/d/1rwDumptgM4jCQQCXWgjeARRHNHkKYRuHJVhwYEFXQRU
> CC [~jieyu] [~jdef] [~karya] [~idownes]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
