[
https://issues.apache.org/jira/browse/SPARK-55993?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-55993:
-----------------------------------
Labels: pull-request-available (was: )
> Flaky RocksDBStateStoreIntegrationSuite: bounded memory usage calculation
> -------------------------------------------------------------------------
>
> Key: SPARK-55993
> URL: https://issues.apache.org/jira/browse/SPARK-55993
> Project: Spark
> Issue Type: Improvement
> Components: Structured Streaming
> Affects Versions: 4.2.0
> Reporter: Kent Yao
> Priority: Major
> Labels: pull-request-available
>
> The test 'bounded memory usage calculation' in
> RocksDBStateStoreIntegrationSuite is flaky due to a shared global state race
> condition in RocksDBMemoryManager (singleton object).
> Line 423 asserts getNumRocksDBInstances(false) == 0 (no unbounded instances),
> but the streaming query may briefly register unbounded instances during
> initialization before the bounded-memory config takes effect. Since
> RocksDBMemoryManager.instanceMemoryMap is a global ConcurrentHashMap shared
> across all tests in the JVM, other concurrent state store operations can also
> register instances.
> The fix should wrap the assertion in eventually to tolerate transient state,
> or move it after processAllAvailable completes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
