Livia Zhu created SPARK-54389:
---------------------------------
Summary: Fix RocksDB State Store Invalid Stamp error when task is
marked as failed during initialization
Key: SPARK-54389
URL: https://issues.apache.org/jira/browse/SPARK-54389
Project: Spark
Issue Type: Improvement
Components: Structured Streaming
Affects Versions: 4.2.0
Reporter: Livia Zhu
Invalid stamp error will be thrown in the following situation:
# Stateful task begins, begins initializing state store and grabs a stamp
# Another task sets {{markTaskFailed}} for the stateful task
# StateStore is initialized,
[setting|https://sourcegraph.prod.databricks-corp.com/databricks-eng/runtime/-/blob/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/RocksDBStateStoreProvider.scala?L192]
the taskCompletion and taskFailure listeners. Since the task was marked as
failed, these listeners are invoked immediately, releasing the stamp. However,
since no failure is thrown, the StateStore initialization completes.
# Stateful task continues to call the {{iterator}} method on the returned
StateStore. Since the stamp is invalid, {{StateStoreErrors.invalidStamp}} is
thrown.
To prevent this error, we should check if the task is failed after setting the
listeners and fail the StateStore initialization if so.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
