[
https://issues.apache.org/jira/browse/FLINK-37524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17937833#comment-17937833
]
Yanfei Lei commented on FLINK-37524:
------------------------------------
[~mayuehappy] Do you mean the differences in
`RocksDBKeyedStateBackend#createState` and
`ForStKeyedStateBackend#createStateInternal`?
The State Serializer is gotten from `tryRegisterKvStateInformation`, if job is
restored from checkpoint,
`kvStateInformation` would be filled by information in checkpoint.
IIUC, ForStKeyedStateBackend implement `tryRegisterKvStateInformation`, so the
old State Serializer in old kvStateInformation would be used. Correct me if I
miss something.
> State Serializer in ForState should restore from oldStateInfo in Checkpoint
> ---------------------------------------------------------------------------
>
> Key: FLINK-37524
> URL: https://issues.apache.org/jira/browse/FLINK-37524
> Project: Flink
> Issue Type: Bug
> Components: Runtime / State Backends
> Affects Versions: 2.0.0
> Reporter: Yue Ma
> Priority: Major
> Fix For: 2.1.0
>
>
> Currently, when creating using ForStKeyedStateBackend create a new State ,
> *_AbstractKeyedState_* will directly use the serializer in
> _*StateDescriptor*_ regardless of whether it is restored from a Checkpoint.
> However, this may cause data to be unable to be restored when the last
> serializer changes (for example, the class registration order in
> *_KryoSerializer_* has changed, for specific cases, please refer to
> {*}_StateBackendTestBase#testKryoRestoreResilienceWithDifferentRegistrationOrder_{*}).
> Therefore, when creating a State from a Checkpoint, oldStateInfo should be
> used as the new State's Serializer.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
