[
https://issues.apache.org/jira/browse/FLINK-3937?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15327451#comment-15327451
]
ASF GitHub Bot commented on FLINK-3937:
---------------------------------------
Github user rmetzger commented on a diff in the pull request:
https://github.com/apache/flink/pull/2085#discussion_r66797580
--- Diff:
flink-yarn/src/main/java/org/apache/flink/yarn/YarnClusterClient.java ---
@@ -211,18 +185,45 @@ public void run() {
Runtime.getRuntime().addShutdownHook(clientShutdownHook);
isConnected = true;
+
+ logAndSysout("Waiting until all TaskManagers have connected");
+
+ while(true) {
--- End diff --
It seems that this loop is now executed also for the YARN session. In the
past, this was used only for "per job" yarn clusters, because there we needed
to have all slots available to submit the job.
Now, this loop is "blocking" the YARN session CLI if not all TaskManager's
are connecting. Its not a big deal, but a change in the semantics.
> Make flink cli list, savepoint, cancel and stop work on Flink-on-YARN clusters
> ------------------------------------------------------------------------------
>
> Key: FLINK-3937
> URL: https://issues.apache.org/jira/browse/FLINK-3937
> Project: Flink
> Issue Type: Improvement
> Reporter: Sebastian Klemke
> Assignee: Maximilian Michels
> Priority: Trivial
> Attachments: improve_flink_cli_yarn_integration.patch
>
>
> Currently, flink cli can't figure out JobManager RPC location for
> Flink-on-YARN clusters. Therefore, list, savepoint, cancel and stop
> subcommands are hard to invoke if you only know the YARN application ID. As
> an improvement, I suggest adding a -yid <yarnApplicationId> option to the
> mentioned subcommands that can be used together with -m yarn-cluster. Flink
> cli would then retrieve JobManager RPC location from YARN ResourceManager.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
