[SPARK-40428][CORE] Fix shutdown hook in the CoarseGrainedSchedulerBackend #37885
Conversation
… since we've got zombie pods hanging around since the resource tie isn't perfect.
dongjoon-hyun
changed the title
| private var _shutdownHookRef: AnyRef = _ | ||
|
|
||
| _shutdownHookRef = ShutdownHookManager.addShutdownHook( | ||
| ShutdownHookManager.DEFAULT_SHUTDOWN_PRIORITY) { () => |
|
SparkContext.stop -> _dagScheduler.stop -> taskScheduler.stop -> backend.stop |
|
hmm for whatever reason that doesn't seem to be triggering for me, let me take a look through and see where things might be getting lost in that call path @mridulm |
…all stop since we've got zombie pods hanging around since the resource tie isn't perfect." This reverts commit dfa8fc5.
…ent of other shutdown failures.
holdenk
changed the title
| starvationTimer.cancel() | ||
| } | ||
| Utils.tryLogNonFatalError { | ||
| abortTimer.cancel() |
There was a problem hiding this comment.
Does Timer.cancel() also require Utils.tryLogNonFatalError?
There was a problem hiding this comment.
Fair point yeah we can back that part out it shouldn't throw unless it's null and it's a val so
|
Looks good to me - modulo @dongjoon-hyun's comment. |
… don't throw anything and add one around fallbackstorage cleanup.
holdenk
changed the title
|
Merged to current dev :) |
3 participants
What changes were proposed in this pull request?
Fix the shutdown hook call through to CoarseGrainedSchedulerBackend
Why are the changes needed?
Sometimes if the driver shuts down abnormally resources may be left dangling.
Does this PR introduce any user-facing change?
No
How was this patch tested?
Existing tests.