[
https://issues.apache.org/jira/browse/FLINK-27273?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17552240#comment-17552240
]
Aitozi edited comment on FLINK-27273 at 6/9/22 2:09 PM:
--------------------------------------------------------
I think it's a valid concern. But we could support other case first. I check
the flink's ha code, can we use
{{HighAvailabilityServicesUtils#createAvailableOrEmbeddedServices}} and do
{{HighAvailabilityServices#closeAndCleanupAllData}}. Or simply clean up the ha
path in the zookeeper
For the job with kerberos enabled we can disable it / validate it now. Or guide
user to enable the operator to access to the zookeeper. (I'm not familiar with
the kerberos, Is this feasible to install the keytab in Flink Kubernetes
Operator?)
was (Author: aitozi):
I think it's a valid concern. But we could support other case first. I check
the flink's ha code, can we use
{{HighAvailabilityServicesUtils#createAvailableOrEmbeddedServices}} and do
{{HighAvailabilityServices#closeAndCleanupAllData}}. Or simply clean up the ha
path in the zookeeper
For the job with kerberos enabled we can disable it / validate it now. Or guide
user to enable the operator to access to the zookeeper. (I'm not familiar with
the kerberos, Is this feasible to install the keytab in Flink Kubernetes
Operator)
> Support both configMap and zookeeper based HA data clean up
> -----------------------------------------------------------
>
> Key: FLINK-27273
> URL: https://issues.apache.org/jira/browse/FLINK-27273
> Project: Flink
> Issue Type: Improvement
> Components: Kubernetes Operator
> Reporter: Aitozi
> Priority: Major
> Fix For: kubernetes-operator-1.1.0
>
>
> As discussed in
> [comments|https://github.com/apache/flink-kubernetes-operator/pull/28#discussion_r815695041]
> We only support clean up the ha data based configMap. Considering that
> zookeeper is still widely used as ha service when deploy on the kubernetes, I
> think we should still take it into account, otherwise, It will come up with
> some unexpected behavior when play with zk ha jobs. For example, it will
> recover old JobGraph when redeploy the application.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
