Dylan Wong created SPARK-60445:
----------------------------------
Summary: Stop the transformWithState state server before a failed
task aborts its state store
Key: SPARK-60445
URL: https://issues.apache.org/jira/browse/SPARK-60445
Project: Spark
Issue Type: Task
Components: Structured Streaming
Affects Versions: 4.3.0
Reporter: Dylan Wong
When a Python {{transformWithState}} task fails or is killed (for example, a
stage abort after an OOM, or a job cancel), the RocksDB store's failure
listener aborts the store, which sets its stamp to -1. The state server is only
stopped by a completion listener, and completion listeners run after failure
listeners. So a Python worker that is still running can make one more state
call, which hits {{{}STATE_STORE_INVALID_STAMP{}}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
