[ 
https://issues.apache.org/jira/browse/FLINK-4356?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16068283#comment-16068283
 ] 

Till Rohrmann commented on FLINK-4356:
--------------------------------------

I think that is an issue for the dispatcher component/application master and 
not the JobManager since it's the former's responsibility to restart failed 
JobManagers. Closing this issue therefore.

> new JobManager HA
> -----------------
>
>                 Key: FLINK-4356
>                 URL: https://issues.apache.org/jira/browse/FLINK-4356
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Cluster Management
>            Reporter: jingzhang
>
> 1. for standalone mode, LocalDispatcher watch JobMaster
> LocalDispatcher detect the failure of JobMaster,  recover jobGraph and 
> Libraries from persistent storage, spawn a new JobManager
> new JobMaster compete for leadership, save address to zookeeper storage
> new JobMaster registers at ResourceManager
> new JobMaster  recover Execution of its job (execution graph) from latest 
> completed checkpoint
> 2. for yarn mode, YarnApplicationMasterRunner create a ProcessReaper of 
> JobMaster
> ProcessReaper monitor JobMaster, kill JVM upon JobMaster termination
> Yarn will create a new AppMaster which contains a new JobManager, JobGraph 
> and Libraries are retrieved as startup artifacts
> new JobMaster compete for leadership, save address to zookeeper storage
> new JobMaster registers at ResourceManager
> new JobMaster  recover Execution of its job (execution graph) from latest 
> completed checkpoint



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

Reply via email to