[
https://issues.apache.org/jira/browse/FLINK-37532?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17937394#comment-17937394
]
Gyula Fora commented on FLINK-37532:
------------------------------------
You need to enable JM HA , this is a limitation of the JM logic not the
operator
> Job recovery always starts from initialSavepointPath when JobManager HA is
> disabled in case of jobManger failure
> -----------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-37532
> URL: https://issues.apache.org/jira/browse/FLINK-37532
> Project: Flink
> Issue Type: Bug
> Components: Kubernetes Operator
> Affects Versions: 1.10.3
> Reporter: Jun He
> Priority: Major
> Labels: kubernetes, operator
>
> When submitting a Flink job using the Flink Kubernetes Operator, if the
> {{initialSavepointPath}} parameter is specified and JobManager High
> Availability (HA) is not enabled, subsequent recovery after the JobManager
> pod crashes for some reason will still restore the job from the path
> specified by {{{}initialSavepointPath{}}}. This leads to the consumption of
> data from a long time ago. Is this a bug in the Kubernetes Operator, or am I
> using the operator incorrectly?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
