Steve Loughran commented on YARN-3417:

For Slider, we are coping with things today (and fixed resource requirements), 
by doing an exit(-1) and restarting. This handles rolling upgrades provided 
requirements don't change and the restart frequency is below that of the 
failure window.

so ... cut it from your todo list for now and we'll see where it ends up.

thanks for offering to do it though —appreciated

> AM to be able to exit with a request saying "restart me with these (possibly 
> updated) resource requirements"
> ------------------------------------------------------------------------------------------------------------
>                 Key: YARN-3417
>                 URL: https://issues.apache.org/jira/browse/YARN-3417
>             Project: Hadoop YARN
>          Issue Type: New Feature
>          Components: resourcemanager
>    Affects Versions: 2.7.0
>            Reporter: Steve Loughran
>            Assignee: Varun Saxena
>            Priority: Minor
> If an AM wants to reconfigure itself or restart with new resources, there's 
> no way to do this without the active participation of a client.
> It can call System.exit and rely on YARN to restart it -but that counts as a 
> failure and may lose the entire app. furthermore, that doesn't allow the AM 
> to resize itself.
> A simple exit-code to be interpreted as restart-without-failure could handle 
> the first case; an explicit call to indicate restart, including potentially 
> new resource/label requirements, could be more reliabile, and certainly more 
> flexible.

This message was sent by Atlassian JIRA

Reply via email to