[
https://issues.apache.org/jira/browse/FLINK-11263?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16733813#comment-16733813
]
Hongtao Zhang commented on FLINK-11263:
---------------------------------------
Hi [~Tison] [~csq]
so currently, the per job mode has the same behavior with the yarn session
other than pre-start the dispatcher and resource manager
the question is that if i want to deploy a job to the cluster.
say the job need 6 parallelism and the each tm only has 4 slots. how does flink
handle this ? does flink will create two TMs since one can't feed the need ?
> yarn session script doesn't boot up task managers
> -------------------------------------------------
>
> Key: FLINK-11263
> URL: https://issues.apache.org/jira/browse/FLINK-11263
> Project: Flink
> Issue Type: Bug
> Components: Cluster Management
> Affects Versions: 1.6.3
> Environment: Flink 1.6.3
> Hadoop 2.7.5
> Reporter: Hongtao Zhang
> Priority: Critical
>
> {{./bin/yarn-session.sh -n 4 -jm 1024m -tm 4096m}}
> {{I want to boot up a Yarn Session Cluster use the command above}}
> {{but Flink doesn't create the task executors but only the
> applicationMaster(YarnSessionClusterEntryPoint)}}
>
> {{the Task Executors will be created when a job was submitted.}}
>
> {{the point is that Yarn ResourceManager doesn't know how many task executors
> should be created because the -n option was not accept by ResourceMananger}}
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
