[
https://issues.apache.org/jira/browse/MESOS-1031?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13930445#comment-13930445
]
Timothy St. Clair commented on MESOS-1031:
------------------------------------------
On [~bmahler]'s suggestion I looked at the patches around signal escalation &
killtree. IMHO it makes more sense to leverage process groups where possible,
and possibly PID namespaces for isolation and ensuring boundaries sans cgroups.
This way the developer can simply call killpg vs. attempting to walk the
processtree. Thoughts?
> Command executor should support signal escalation.
> --------------------------------------------------
>
> Key: MESOS-1031
> URL: https://issues.apache.org/jira/browse/MESOS-1031
> Project: Mesos
> Issue Type: Improvement
> Components: slave
> Reporter: Niklas Quarfot Nielsen
>
> To make command executor shutdown more graceful, the sequence could allow
> signal escalation with initial SIGTERM followed by SIGKILL after an
> appropriate timeout. This became apparent when we where using the command
> executor for pluggable containerization (see
> https://issues.apache.org/jira/browse/MESOS-816).
> The escalation would need to prevent shutdown of the command executor until
> child processes have actually stopped and if processes are still alive after
> the timeout, send SIGKILL to the original process group.
> Further more, additional unit tests could be added to verify the new
> escalation i.e. add or modify a test-executor so it becomes irresponsive to
> SIGTERM and/or leaves behind orphan processes.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
