Dian Fu created FLINK-21916:
-------------------------------
Summary: Allows multiple kinds of ManagedMemoryUseCase for the
same operator
Key: FLINK-21916
URL: https://issues.apache.org/jira/browse/FLINK-21916
Project: Flink
Issue Type: Sub-task
Components: API / DataStream
Reporter: Dian Fu
Assignee: Dian Fu
Fix For: 1.13.0
Currently, the implementation of
[SimpleTransformationTranslator|https://github.com/apache/flink/blob/97bfd049951f8d52a2e0aed14265074c4255ead0/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/graph/SimpleTransformationTranslator.java#L119]
restricts that OPERATOR and PYTHON use case could not be used in the same
operator. For batch mode of Python DataStream API:
- OPERATOR use case will be added firstly in
BatchExecutionUtils.applyBatchExecutionSettings for a keyed operator during
translation
- Then the PYTHON use case which is set in the transformation could not be
added in
[SimpleTransformationTranslator|https://github.com/apache/flink/blob/97bfd049951f8d52a2e0aed14265074c4255ead0/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/graph/SimpleTransformationTranslator.java#L119]
as the streamNode.getManagedMemoryOperatorScopeUseCaseWeights() is not empty.
We need to remove the restriction in
[SimpleTransformationTranslator|https://github.com/apache/flink/blob/97bfd049951f8d52a2e0aed14265074c4255ead0/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/graph/SimpleTransformationTranslator.java#L119]
if there are no special reasons.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
