[SPARK-18761][branch-2.0] Introduce "task reaper" to oversee task killing in executors #16358
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…n executors Spark's current task cancellation / task killing mechanism is "best effort" because some tasks may not be interruptible or may not respond to their "killed" flags being set. If a significant fraction of a cluster's task slots are occupied by tasks that have been marked as killed but remain running then this can lead to a situation where new jobs and tasks are starved of resources that are being used by these zombie tasks. This patch aims to address this problem by adding a "task reaper" mechanism to executors. At a high-level, task killing now launches a new thread which attempts to kill the task and then watches the task and periodically checks whether it has been killed. The TaskReaper will periodically re-attempt to call `TaskRunner.kill()` and will log warnings if the task keeps running. I modified TaskRunner to rename its thread at the start of the task, allowing TaskReaper to take a thread dump and filter it in order to log stacktraces from the exact task thread that we are waiting to finish. If the task has not stopped after a configurable timeout then the TaskReaper will throw an exception to trigger executor JVM death, thereby forcibly freeing any resources consumed by the zombie tasks. This feature is flagged off by default and is controlled by four new configurations under the `spark.task.reaper.*` namespace. See the updated `configuration.md` doc for details. Tested via a new test case in `JobCancellationSuite`, plus manual testing. Author: Josh Rosen <joshrosen@databricks.com> Closes apache#16189 from JoshRosen/cancellation.
|
Test build #70419 has finished for PR 16358 at commit
|
Contributor
|
LGTM. Merging to branch-2.0 |
asfgit
pushed a commit
that referenced
this pull request
Dec 20, 2016
…ling in executors Branch-2.0 backport of #16189; original description follows: ## What changes were proposed in this pull request? Spark's current task cancellation / task killing mechanism is "best effort" because some tasks may not be interruptible or may not respond to their "killed" flags being set. If a significant fraction of a cluster's task slots are occupied by tasks that have been marked as killed but remain running then this can lead to a situation where new jobs and tasks are starved of resources that are being used by these zombie tasks. This patch aims to address this problem by adding a "task reaper" mechanism to executors. At a high-level, task killing now launches a new thread which attempts to kill the task and then watches the task and periodically checks whether it has been killed. The TaskReaper will periodically re-attempt to call `TaskRunner.kill()` and will log warnings if the task keeps running. I modified TaskRunner to rename its thread at the start of the task, allowing TaskReaper to take a thread dump and filter it in order to log stacktraces from the exact task thread that we are waiting to finish. If the task has not stopped after a configurable timeout then the TaskReaper will throw an exception to trigger executor JVM death, thereby forcibly freeing any resources consumed by the zombie tasks. This feature is flagged off by default and is controlled by four new configurations under the `spark.task.reaper.*` namespace. See the updated `configuration.md` doc for details. ## How was this patch tested? Tested via a new test case in `JobCancellationSuite`, plus manual testing. Author: Josh Rosen <joshrosen@databricks.com> Closes #16358 from JoshRosen/cancellation-branch-2.0.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Branch-2.0 backport of #16189; original description follows:
What changes were proposed in this pull request?
Spark's current task cancellation / task killing mechanism is "best effort" because some tasks may not be interruptible or may not respond to their "killed" flags being set. If a significant fraction of a cluster's task slots are occupied by tasks that have been marked as killed but remain running then this can lead to a situation where new jobs and tasks are starved of resources that are being used by these zombie tasks.
This patch aims to address this problem by adding a "task reaper" mechanism to executors. At a high-level, task killing now launches a new thread which attempts to kill the task and then watches the task and periodically checks whether it has been killed. The TaskReaper will periodically re-attempt to call
TaskRunner.kill()and will log warnings if the task keeps running. I modified TaskRunner to rename its thread at the start of the task, allowing TaskReaper to take a thread dump and filter it in order to log stacktraces from the exact task thread that we are waiting to finish. If the task has not stopped after a configurable timeout then the TaskReaper will throw an exception to trigger executor JVM death, thereby forcibly freeing any resources consumed by the zombie tasks.This feature is flagged off by default and is controlled by four new configurations under the
spark.task.reaper.*namespace. See the updatedconfiguration.mddoc for details.How was this patch tested?
Tested via a new test case in
JobCancellationSuite, plus manual testing.