[
https://issues.apache.org/jira/browse/FLINK-26179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17500532#comment-17500532
]
Yun Tang commented on FLINK-26179:
----------------------------------
I don't think Flink community should support automatic savepoints in Flink k8s
operator currently.
>From my understanding, the k8s operator, which made by Google, introduced the
>automatic savepoints for the purpose of [auto
>failover|https://github.com/GoogleCloudPlatform/flink-on-k8s-operator/blob/master/docs/savepoints_guide.md#automatically-restarting-job-from-the-latest-savepoint].
> Since we already support high-availiabty and could restore from LAST_STATE, I
>cannot see any beneficts compared with period checkpoints.
Current default savepoint format type is still CANONICAL which will scan the
whole DB to write all the key-value pairs one by one. This is extremely slow
and expensive, not to mentition the addition memory usage and additional disk
usage to DFS in the large scale case.
> Support periodic savepointing in the operator
> ---------------------------------------------
>
> Key: FLINK-26179
> URL: https://issues.apache.org/jira/browse/FLINK-26179
> Project: Flink
> Issue Type: Sub-task
> Components: Kubernetes Operator
> Reporter: Gyula Fora
> Priority: Major
>
> Automatic triggering of savepoints is a commonly requested feature. The
> configuration should be part of the job spec.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
