jtuglu1 opened a new issue, #18464:
URL: https://github.com/apache/druid/issues/18464

   Ghost locks can be allocated due to races between `TaskQueue` and 
`TaskLockbox`. These can stall future kill/compaction tasks and/or other tasks 
which use locks.
   
   ### Affected Version
   
   Druid v34, although I suspect it's earlier versions too.
   
   ### Description
   [T=1,Thread=1]: Overlord shutdown task A, executing 
[this](https://github.com/apache/druid/blob/-/indexing-service/src/main/java/org/apache/druid/indexing/overlord/TaskQueue.java#L744)
 line.
   [T=1,Thread=2]: Task begins 
[running](https://github.com/apache/druid/blob/-/indexing-service/src/main/java/org/apache/druid/indexing/overlord/TaskLockbox.java#L485)
 a SegmentAllocate action
   [T=1,Thread=2]: 
[This](https://github.com/apache/druid/blob/-/indexing-service/src/main/java/org/apache/druid/indexing/overlord/TaskLockbox.java#L498)
 line executes, returning a non-empty `AllocationHolderList` because indeed the 
task is still in `activeTasks`.
   [T=1,Thread=2]: Releases `giant` lock.
   [T=1,Thread=1]: Acquires `giant` lock and 
[removes](https://github.com/apache/druid/blob/-/indexing-service/src/main/java/org/apache/druid/indexing/overlord/TaskLockbox.java#L1239)
 it from the `activeTasks` set.
   Result: We receive the shutdown signal, but we allocate a lock for the task 
A (or do really any stateful operation on the overlord) before actually 
shutting down the task (race).
   
   


-- 
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]

Reply via email to