maoyuwei1 opened a new issue, #13279: URL: https://github.com/apache/gluten/issues/13279
### Backend VL (Velox) ### Bug description CelebornUtils.fetchCleanFailedShuffle(lifecycleManager, celebornConf, failedShuffleCleaner) and stageRerun(...) take failedShuffleCleaner as an output parameter and assign the newly created FailedShuffleCleaner instance to it. Because Java is pass-by-value, reassigning a method parameter is not visible to the caller — CelebornShuffleManager.failedShuffleCleaner stays null forever. org.apache.spark.shuffle.gluten.celeborn.CelebornShuffleManager <img width="1351" height="634" alt="Image" src="https://github.com/user-attachments/assets/68fd2142-dfca-4823-91e1-b9ab6fe94214"; /> <img width="783" height="706" alt="Image" src="https://github.com/user-attachments/assets/ff02f63c-0f31-483a-b0c1-543f470cd34f"; /> becomes dead code. The cleaner's internal ScheduledExecutorService is never shut down, and keeps strong references to LifecycleManager and its pending shuffle-id collections, leaking both threads and objects. This accumulates across SparkContext lifecycles and is most visible on long-lived drivers (e.g., Spark Thrrift Server), where each restart leaves orphaned cleaner threads and repeated invalid clean log entries. Expected behavior stageRerun/fetchCleanFailedShuffle should return the created instance so the caller can store it and stop() can release it. ### Gluten version _No response_ ### Spark version None ### Spark configurations _No response_ ### System information _No response_ ### Relevant logs ```bash ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
