[jira] [Commented] (FLINK-7463) Release nascent workers on slot request timeout

["Till Rohrmann (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-7463?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16134130#comment-16134130 ] 

Till Rohrmann commented on FLINK-7463:
--------------------------------------

I think this is not only a subtask of the Flink on Mesos integration but applies to Yarn as well. Maybe we can convert this into an issue.

> Release nascent workers on slot request timeout 
> ------------------------------------------------
>
>                 Key: FLINK-7463
>                 URL: https://issues.apache.org/jira/browse/FLINK-7463
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Mesos
>            Reporter: Eron Wright 
>
> A slot request causes a new worker to be allocated.   If the slot request times out or is cancelled before the worker is launched, cancel the worker request if possible.
> This is an optimization because an idle slot is eventually released anyway.   However, I observe that a lot of worker requests pile up in the launch coordinator, as the JM keeps making request after request.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)