Github user tillrohrmann commented on the issue:
https://github.com/apache/flink/pull/2449
Thanks for the quick and thorough review @StephanEwen.
- Concerning the `ExecutionContext` in the `TaskManager`: This is simply
the `ExecutionContext` which was passed before to the `NetworkEnvironment` but
now is no longer needed there. I agree that we should be careful with thread
pool creations. At the moment, we create for each `JobManager` and
`TaskManager` an additional execution context. The advantage of this approach
is that we separate the actor execution from the execution of other components.
Thus, it's not possible that another component starves the execution context of
the actor and thus influences its responsiveness. But I agree that the few
additional future callbacks shouldn't be a big burden for the `TaskManager`
actor system. Will change it.
- It is true that there is no notion of slots on the `TaskManager` right
now. I've named it `SlotEnvironment` because with the Flip-6 refactorings we
need to maintain multiple job manager connections and every task runs in a slot
which is associated with a JobManager. Renaming it to `JobManagerConnection`
should better reflect what it actually is.
- Yes that is a good idea. Will try to attach the
`ResultPartitionConsumableNotifier` to the `Task`.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
