[
https://issues.apache.org/jira/browse/FLINK-29913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17725286#comment-17725286
]
Feifan Wang commented on FLINK-29913:
-------------------------------------
Sorry I didn't notice this ticket earlier so I submitted a duplicate one.
[~Yanfei Lei] for the priority, I agree with [~klion26] , since it may break
the checkpoint in valid use case, the priority at least be *Major* .
[~roman] I think your proposal is valid, but I still want to provide an
alternative :
# Make SharedStateRegistry allow register multi state object to same key. The
generation strategy for SharedStateRegistryKey is still determined by
CompositeStateHandle , as it is now. Different state objects may be registered
under the same key in force-full-checkpoint (active re-upload).
# SharedStateRegistry maintains an linked entry list sorted by registration
time for each key. PlaceHolderStreamStateHandle will be replace by the last one
in the entry list.
In this approach, SharedStateRegistry don't discard any state object in
registration process. SharedStateRegistry only deleted state object when its
lastUsedCheckpoint is subsumed.
> Shared state would be discarded by mistake when maxConcurrentCheckpoint>1
> -------------------------------------------------------------------------
>
> Key: FLINK-29913
> URL: https://issues.apache.org/jira/browse/FLINK-29913
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Checkpointing
> Affects Versions: 1.15.0, 1.16.0
> Reporter: Yanfei Lei
> Priority: Minor
>
> When maxConcurrentCheckpoint>1, the shared state of Incremental rocksdb state
> backend would be discarded by registering the same name handle. See
> [https://github.com/apache/flink/pull/21050#discussion_r1011061072]
> cc [~roman]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
