[
https://issues.apache.org/jira/browse/YARN-6305?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15901655#comment-15901655
]
Shane Kumpf commented on YARN-6305:
-----------------------------------
I've been looking into this, so I'll take ownership and will put together a
patch for discussion.
> Improve signaling of short lived containers
> -------------------------------------------
>
> Key: YARN-6305
> URL: https://issues.apache.org/jira/browse/YARN-6305
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: yarn
> Reporter: Shane Kumpf
>
> Currently it is possible for containers to leak and remain in an exited state
> if a docker container is not fully started before being killed. Depending on
> the selected Docker storage driver, the lower bound on starting a container
> can be as much as three seconds (using {{docker run}}). If an implicit image
> pull occurs, this could be much longer.
> When a container is not fully started, the PID is not available yet. As a
> result, {{ContainerLaunch#cleanUpContainer}} will not signal the container as
> it relies on the PID. The PID is not required for docker client operations,
> so allowing the signaling to occur anyway appears to be appropriate.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
