[
https://issues.apache.org/jira/browse/FLINK-21935?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17307562#comment-17307562
]
Yu Li commented on FLINK-21935:
-------------------------------
+1 to remove the option and synchronous snapshot support for heap backend.
Shall we send a notice in our user mailing list or release note is enough?
Since the default value of {{state.backend.async}} is {{true}}, I believe the
use of synchronous snapshot is pretty rare.
I could see more related works to do, such as deprecating
{{NestedMapsStateTable}} and do clean ups in the next release (or maybe we
could carry this out in this release since if we remove the option then no way
to use it after all?).
I will be ready to do the PR review, just let me know (smile)
> Remove "state.backend.async" option.
> ------------------------------------
>
> Key: FLINK-21935
> URL: https://issues.apache.org/jira/browse/FLINK-21935
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / State Backends
> Reporter: Stephan Ewen
> Priority: Blocker
> Fix For: 1.13.0
>
>
> Checkpoints are always asynchronous, there is no case ever for a synchronous
> checkpoint.
> The RocksDB state backend doesn't even support synchronous snapshots, and the
> HashMap Heap backend also has no good use case for synchronous snapshots
> (other than a very minor reduction in heap objects).
> Most importantly, we should not expose this option in the constructors of the
> new state backend API classes, like {{HashMapStateBackend}}.
> I marked this a blocker because it is part of the new user-facing State
> Backend API and I would like to avoid that this option enters this API and
> causes confusion when we eventually remove it.
> /cc [~sjwiesman] and [~liyu]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
