[
https://issues.apache.org/jira/browse/FLINK-18211?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17138968#comment-17138968
]
Yang Wang commented on FLINK-18211:
-----------------------------------
FLINK-13837 is not only for Yarn deployment. It could be used for all session
cluster(including standalone, Yarn, K8s). The artifacts and jars will be
shipped to jobmanager via rest client and then put into blob storage. It is
completely different from Yarn ship mechanism(via Yarn local resources).
For Yarn session cluster, it is also not redundant. Since we could specify
dependencies for each job to avoid conflicts. But i admin that it may be
redundant for per-job cluster.
The {{env.registerCachedFile}} is good enough for user uber jar. However, if
he/she has some dependencies(e.g. jars, config files) to ship, he/she has to
register them in codes. It is not flexible, especially for user config files. I
think that is also why users want {{pipeline.jars}} could be specified via
users.
> Dynamic properties setting 'pipeline.jars' will be overwritten
> --------------------------------------------------------------
>
> Key: FLINK-18211
> URL: https://issues.apache.org/jira/browse/FLINK-18211
> Project: Flink
> Issue Type: Bug
> Components: Client / Job Submission
> Affects Versions: 1.10.0, 1.11.0
> Reporter: Echo Lee
> Assignee: Echo Lee
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.12.0
>
>
> When we submit the application through "flink run
> -Dpipeline.jars='/user1.jar, user2.jar'..." command, configuration will
> include 'pipeline.jars', But ExecutionConfigAccessor#fromProgramOptions will
> be reset this property, So the property set by the user is invalid.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
