[
https://issues.apache.org/jira/browse/FLINK-6374?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15994438#comment-15994438
]
Till Rohrmann commented on FLINK-6374:
--------------------------------------
I think this is a convenience function for users who specify their clusters
wrongly. It does not make sense to start a cluster for a job whose parallelism
exceeds the number of available slots. Thus, {{-p}} is overriding {{-ys}} if
{{-p}} is larger than {{-ys}}. Therefore, I think we can close this issue.
> In yarn cluster model, the -ys parameter is not take effect.
> ------------------------------------------------------------
>
> Key: FLINK-6374
> URL: https://issues.apache.org/jira/browse/FLINK-6374
> Project: Flink
> Issue Type: Bug
> Components: YARN
> Reporter: zhangrucong1982
>
> I use the flink version of 1.2.0. I deploy the Flink cluster on the yarn. I
> have three nodes. Every node has 32GB memory and 16 slots. And the flink
> key configuration is :yarn.containers.vcores: 16.
> I use two yarn deployment models, but the result is not same.The details are
> the following:
> 1、First I use the flink yarn session model:
> 1) I start the flink yarn cluster with the command:./yarn-session.sh -n 2 -s
> 1 -d .
> 2)Then I submit a streaming job with the command : ./flink run -p 3
> ../examples/streaming/WindowJoin.jar ;
> 3) the streaming job deploy failure. The reason is:"Not enough free slots
> available to run the job. You can decrease the operator parallelism or
> increase the number of slots per TaskManager in the configuration."
> 2、The I use the yarn cluster model:
> 1) I start the yarn cluster and submit a streaming job with command " ./flink
> run -m yarn-cluster -yn 2 -ys 1 -p 3 ../examples/streaming/WindowJoin.jar "
> but, the streaming job is runing;
> 2) I look at the code, if there is a -p parameter in the command, the -ys
> parameter is not take effect.
> what is the design for -ys parameter? I think the result of yarn session is
> reasonable, the yarn cluster is not reasonable. How do you think?
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
