[
https://issues.apache.org/jira/browse/FLINK-33096?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17767595#comment-17767595
]
wawa commented on FLINK-33096:
------------------------------
So sorry，correct the previous description:
there was an exception during the scheduling of the taskManager, and the
restart strategy took effect. As a result, the Duration value was reset and
started counting again.
However, due to pod memory overuse, the taskManager pod was killed and evicted
by Kubernetes. When trying to schedule a new taskManager pod, it failed,
resulting in the failure of the entire Flink job.
> Flink on k8s，if one taskmanager pod was crashed，the whole flink job will be
> failed
> ----------------------------------------------------------------------------------
>
> Key: FLINK-33096
> URL: https://issues.apache.org/jira/browse/FLINK-33096
> Project: Flink
> Issue Type: Bug
> Components: Deployment / Kubernetes
> Affects Versions: 1.14.3
> Reporter: wawa
> Priority: Major
>
> The Flink version is 1.14.3, and the job is submitted to Kubernetes using the
> Native Kubernetes application mode. During the scheduling process, when a
> TaskManager pod crashes due to an exception, Kubernetes will attempt to start
> a new TaskManager pod. However, the scheduling process is halted immediately,
> resulting in the entire Flink job being terminated. On the other hand, if the
> JobManager pod crashes, Kubernetes is able to successfully schedule a new
> JobManager pod. This observation was made during application usage. Can you
> please help analyze the underlying issue?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
