You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Till Rohrmann (JIRA)" <ji...@apache.org> on 2018/07/13 07:26:00 UTC

[jira] [Commented] (FLINK-7470) Acquire RM leadership before registering with Mesos

    [ https://issues.apache.org/jira/browse/FLINK-7470?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16542649#comment-16542649 ] 

Till Rohrmann commented on FLINK-7470:
--------------------------------------

The {{ResourceManager}} has now a leader election service. We could leverage the {{ResourceManager#grantLeadership}} callback to trigger the registration of the framework at the Mesos master.

> Acquire RM leadership before registering with Mesos
> ---------------------------------------------------
>
>                 Key: FLINK-7470
>                 URL: https://issues.apache.org/jira/browse/FLINK-7470
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Mesos
>            Reporter: Eron Wright 
>            Priority: Major
>             Fix For: 1.7.0
>
>
> Mesos doesn't support fencing tokens in the scheduler protocol; it assumes external leader election among scheduler instances.   The last connection wins; prior connections for a given framework ID are closed.
> The Mesos RM should not register as a framework until it has acquired RM leadership.   Evolve the ResourceManager as necessary.   One option is to introduce an ResourceManagerRunner that acquires leadership before starting the RM.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)