GitHub user StephanEwen opened a pull request:
https://github.com/apache/flink/pull/5658
[FLINK-8856] [TaskManager] Move all cancellation interrupt calls to
TaskCanceller thread
## What is the purpose of the change
This cleans up the code and guards against a JVM bug where `interrupt()`
calls
block/deadlock if the thread is engaged in certain I/O operations.
In addition, this makes sure that the process really goes away when the
cancellation
timeout expires, rather than relying on the TaskManager to be able to
properly handle
the fatal error notification.
Some minor robustness enhancements related to this change are included in
this PR.
## Verifying this change
The change is motivated by an occasional JVM bug that I could not
purposefully trigger in tests to guard against rollback to the prior state. All
tests were passing prior to this change and are passing after this change.
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): (yes / **no**)
- The public API, i.e., is any changed class annotated with
`@Public(Evolving)`: (yes / **no**)
- The serializers: (yes / **no** / don't know)
- The runtime per-record code paths (performance sensitive): (yes /
**no** / don't know)
- Anything that affects deployment or recovery: JobManager (and its
components), Checkpointing, Yarn/Mesos, ZooKeeper: (**yes** / no / don't know)
- The S3 file system connector: (yes / **no** / don't know)
## Documentation
- Does this pull request introduce a new feature? (yes / **no**)
- If yes, how is the feature documented? (**not applicable** / docs /
JavaDocs / not documented)
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/StephanEwen/incubator-flink fix_task_interrupt
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/5658.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #5658
commit 36726fc5c277c649dc360975a34bf7be0afd7a0e
Author: Stephan Ewen
Date: 2018-03-06T14:54:13Z
[hotfix] [taskmanager] Fix checkstyle in Task and TaskTest
commit 385b2032bcaa397d21d28e90efa89a44e12ebe99
Author: Stephan Ewen
Date: 2018-03-06T14:18:33Z
[FLINK-8856] [TaskManager] Move all cancellation interrupt calls to
TaskCanceller thread
This cleans up the code and guards against a JVM bug where 'interrupt()'
calls
block/deadlock if the thread is engaged in certain I/O operations.
In addition, this makes sure that the process really goes away when the
cancellation
timeout expires, rather than relying on the TaskManager to be able to
properly handle
the fatal error notification.
commit 3b18d7d9eccc936dc53b05b787f3fd4c19171d4f
Author: Stephan Ewen
Date: 2018-03-06T15:36:13Z
[FLINK-8883] [core] Make ThreadDeath a fatal error in ExceptionUtils
commit f3884088b210a061dba4d83323884bece1d31864
Author: Stephan Ewen
Date: 2018-03-06T16:14:54Z
[FLINK-8885] [TaskManager] DispatcherThreadFactory registers a fatal error
exception handler
In case dispatcher threads let an exception bubble out (does not hanle it),
the
exception handler terminates the process, to esure we don't leave broken
processes.
commit 5236bb73a2482ccdf016d1b9bea5cd0f17f2f620
Author: Stephan Ewen
Date: 2018-03-06T16:18:38Z
[hotfix] [runtime] Harden FatalExitExceptionHandler
In case the logging framework throws an exception when handling the
exception,
we still kill the process, as intended.
---