You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@slider.apache.org by "Jonathan Maron (JIRA)" <ji...@apache.org> on 2014/06/02 21:01:04 UTC

[jira] [Commented] (SLIDER-85) Slider AM to recognise & fail when another instance of the same name is up.

    [ https://issues.apache.org/jira/browse/SLIDER-85?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14015740#comment-14015740 ] 

Jonathan Maron commented on SLIDER-85:
--------------------------------------

is the idea to have the AM test whether to proceed with a launch, assuming that perhaps an instance of the same name was launched between the launch request submitted by the slider client and the AM launch?  The client already has some "clusterExists" checks based on checking the cluster dir.

> Slider AM to recognise & fail when another instance of the same name is up.
> ---------------------------------------------------------------------------
>
>                 Key: SLIDER-85
>                 URL: https://issues.apache.org/jira/browse/SLIDER-85
>             Project: Slider
>          Issue Type: Improvement
>          Components: appmaster, registry
>    Affects Versions: Slider 0.30
>            Reporter: Steve Loughran
>   Original Estimate: 3h
>  Remaining Estimate: 3h
>
> using the YARN app reports to enumerate instances before launching has a race condition two attempts to deploy the same application instance could be made. If neither operation was visible in the queue when it was polled, they'd both get in
> All HBase/accumulo instances would join up to form an aggregate cluster, which could be confusing, and complicate other operations like "stop".
> The registry can be used to fix this: if there is an instance already of that name, the AM should fail



--
This message was sent by Atlassian JIRA
(v6.2#6252)