[FLINK-35166][runtime] Make the SortBufferAccumulator use more buffers when the parallelism is small

[jiangxin369]

What is the purpose of the change

Improve the performance of hybrid shuffle when enable memory decoupling and meantime the parallelism is small.

Brief change log

• Make the SortBufferAccumulator use more buffers when the parallelism is small

Verifying this change

Please make sure both new and modified tests in this PR follow the conventions for tests defined in our code quality guide.

This change is a trivial rework / code cleanup without any test coverage.

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): (no)
• 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)

[flinkbot]

CI report:

• dbb93da Azure: SUCCESS

Bot commands

The @flinkbot bot supports the following commands:

• @flinkbot run azure re-run the last Azure build

[TanYuxin-tyx]

Thanks @jiangxin369 , LGTM.

[reswqa]

Is TPC-DS performance no longer regression after the fix?

[jiangxin369]

Is TPC-DS performance no longer regression after the fix?

@reswqa The regression still exists because we replace the HashBufferAccumulator with the SortBufferAccumulator when the decoupling is enabled and the parallelism is less than 512, but this PR reduces the regression. According to the previous discussion, the regression is acceptable if the feature is enabled.

[reswqa]

Is TPC-DS performance no longer regression after the fix?

@reswqa The regression still exists because we replace the HashBufferAccumulator with the SortBufferAccumulator when the decoupling is enabled and the parallelism is less than 512, but this PR reduces the regression. According to the previous discussion, the regression is acceptable if the feature is enabled.

Ah, I see. Let's keep improving it then :)