[
https://issues.apache.org/jira/browse/FLINK-4156?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15365978#comment-15365978
]
Robert Metzger commented on FLINK-4156:
---------------------------------------
[~aljoscha]: This only happens if the same zookeeper namespace has been
configured for both jobs.
This can easily happen if you put the zookeeper namespace into the
conf/flink-conf.yaml file, because then both the YARN session and the per-job
yarn cluster use the same config.
But if users put different namespaces into the config between the jobs (or they
pass them using ./bin/yarn-session.sh -Dzk.foo=bar) then it will work properly.
As Stefan said, an easy workaround would be setting a UUID by default in the
YARN code.
> Job with -m yarn-cluster registers TaskManagers to another running Yarn
> session
> -------------------------------------------------------------------------------
>
> Key: FLINK-4156
> URL: https://issues.apache.org/jira/browse/FLINK-4156
> Project: Flink
> Issue Type: Bug
> Components: Distributed Coordination, YARN Client
> Reporter: Stefan Richter
>
> When a job is started using cluster mode (-m yarn-cluster) and a Yarn session
> is running on the same cluster, the job accidentally registers it's worker
> tasks with the ongoing Yarn session. This happens because the same Zookeeper
> namespace is used.
> We should consider isolating Flink applications from another by using UUIDS,
> e.g. based on their application ids, in their Zookeeper paths.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
