[
https://issues.apache.org/jira/browse/FLINK-20206?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17235299#comment-17235299
]
Yang Wang commented on FLINK-20206:
-----------------------------------
Currently, when the Flink client creates the JobManager deployment
successfully, it will exit normally. However, the JobManager may not be started
and crash in many situation(e.g. wrong image, service account is not correctly
set, JobManager internal error, etc.). Then the users need to use {{kubectl
describe pod <podName>}} or {{kubectl logs <podName>}} to figure out the root
cause.
I am not sure whether it is a good solution to let the Flink client always
waiting for JobManager launched. Since it will take long time if the image has
not been pulled ever. This could make the QPS of deployer very low and then it
will be a bottleneck.
I will keep this ticket open and get more suggestions from the users.
> Failed to start the session, but there is no clear prompt.
> ----------------------------------------------------------
>
> Key: FLINK-20206
> URL: https://issues.apache.org/jira/browse/FLINK-20206
> Project: Flink
> Issue Type: Improvement
> Components: Deployment / Kubernetes
> Affects Versions: 1.12.0
> Reporter: Guowei Ma
> Priority: Critical
> Attachments: image-2020-11-18-15-12-13-530.png
>
>
> Use ./bin/kubernetes-session.sh to start a k8s session clustter. The log
> showes the session cluster successfully start but it not.
> Personally I prefer the yarn-session way, which could make me have a clear
> expectation.
> So I would like to propose that Flink could give more detail information
> about whether session cluster create success or not.
> !image-2020-11-18-15-12-13-530.png!
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
