Yun Tang created FLINK-25479:
--------------------------------
Summary: Changlog materialization with incremental checkpoint
cannot work well in local tests
Key: FLINK-25479
URL: https://issues.apache.org/jira/browse/FLINK-25479
Project: Flink
Issue Type: Bug
Components: Runtime / Checkpointing, Runtime / State Backends, Tests
Reporter: Yun Tang
Fix For: 1.15.0
Currently, changelog materialization would call RocksDB state backend's
snapshot method to generate {{IncrementalRemoteKeyedStateHandle}} as
ChangelogStateBackendHandleImpl's materialized artifacts. And before next
materialization, it will always report the same
{{IncrementalRemoteKeyedStateHandle}} as before.
For local tests, TM would report the {{IncrementalRemoteKeyedStateHandle}} to
JM via local {{LocalRpcInvocation}}. However, as {{LocalRpcInvocation}} would
not de/serialize message, which leads once we register the
{{IncrementalRemoteKeyedStateHandle}} on JM side, it will also add a
{{sharedStateRegistry}} to the one located on TM side. For the 2nd checkpoint,
TM would reported same {{IncrementalRemoteKeyedStateHandle}} with
{{sharedStateRegistry}} to JM. And it will then throw exception as it already
contains a {{sharedStateRegistry}}:
IncrementalRemoteKeyedStateHandle
{code:java}
public void registerSharedStates(SharedStateRegistry stateRegistry, long
checkpointID) {
Preconditions.checkState(
sharedStateRegistry != stateRegistry,
"The state handle has already registered its shared states to
the given registry.");
}
{code}
This bug would go in distribution environment as
{{IncrementalRemoteKeyedStateHandle}} would be serialized and
{{sharedStateRegistry}} is tagged as {{transient}}.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
