[
https://issues.apache.org/jira/browse/SPARK-56496?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Xingbo Jiang resolved SPARK-56496.
----------------------------------
Resolution: Fixed
Resolved by https://github.com/apache/spark/pull/55360
> maxConsecutiveStageAttempts vs maxStageAttempts dual-limit inconsistency
> ------------------------------------------------------------------------
>
> Key: SPARK-56496
> URL: https://issues.apache.org/jira/browse/SPARK-56496
> Project: Spark
> Issue Type: Sub-task
> Components: Spark Core
> Affects Versions: 4.1.1
> Reporter: Xingbo Jiang
> Assignee: Xingbo Jiang
> Priority: Major
> Labels: pull-request-available
>
> *Problem*
> DAGScheduler has two separate stage-retry limits: maxConsecutiveStageAttempts
> (for consecutive failures) and maxStageAttempts (total attempts). Different
> code paths check only one of the two limits, so the effective abort behavior
> depends on which path is taken. This can cause unexpected job aborts
> (stricter limit wins silently) or prevent expected aborts.
> *Impact*
> Unpredictable job abort behavior. Users configuring one limit may be
> surprised by the other. Inconsistent behavior between fetch-failure paths and
> non-fetch-failure paths.
> *Suggested Fix*
> Document the exact interaction (both limits are enforced independently;
> whichever triggers first aborts the stage). Unify the limit check into a
> single canRetryStage helper used by all paths.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
