Zakelly opened a new pull request, #25949:
URL: https://github.com/apache/flink/pull/25949
## What is the purpose of the change
Currently, there is a limit for in-flight records in async state processing.
However, we allow developers to initialize a new processing in-middle of
another. These derived processings are treated as normal ones for timers or
upstreaming records, which is problematic:
- The new derived processing may increase the number of in-flight requests
and cause force draining when it reaches the limit. But the draining may
require the current processing also finish. This is a deadlock.
- The derived processing will queue behind normal processing, which is not
the behavior the user wants. The derived ones should be fired right after the
current processing or ASAP.
Thus, This PR changes the behavior in `AsyncExecutionController` and its
related buffer:
- Avoid drain when creating derived processing.
- Provide priority for the queuing processing and make derived one have
greater priority.
## Brief change log
- Add an additional condition to skip the drain when inserting new record
to `AEC`'s buffer.
- Add priority to `RecordContext` and consider it when queue in blocking
queue.
## Verifying this change
This change added tests and can be verified by
`AbstractAsyncStateStreamOperatorTest#testManyAsyncProcessWithKey`
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): no
- The public API, i.e., is any changed class annotated with
`@Public(Evolving)`: no
- The serializers: no
- The runtime per-record code paths (performance sensitive): yes
- Anything that affects deployment or recovery: JobManager (and its
components), Checkpointing, Kubernetes/Yarn, ZooKeeper: no
- The S3 file system connector: no
## Documentation
- Does this pull request introduce a new feature? no
- If yes, how is the feature documented? not applicable
--
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]
