zhijiang created FLINK-13010:
--------------------------------
Summary: Refactor the process of SchedulerNG#requestPartitionState
Key: FLINK-13010
URL: https://issues.apache.org/jira/browse/FLINK-13010
Project: Flink
Issue Type: Sub-task
Components: Runtime / Coordination, Runtime / Network
Reporter: zhijiang
Assignee: zhijiang
Currently `requestPartitionState` is mainly used for querying partition state
when the consumer receives `PartitionNotFoundException` during requesting
partition. Actually we do not have the concept of partition state atm, and `
requestPartitionState` would return the corresponding producer's state as as
result, so it exists a contradiction here.
My suggestion is refactoring the method as `requestPartitionProducerState` and
we do not need to pass `IntermediateDataSetID` and `ResultPartitionID`
arguments for finding the corresponding execution attempt. We could only pass
the `ExecutionAttemptID` in method then the corresponding execution attempt
could be easily found from the mapping in `ExecutionGraph`.
To do so, we could further remove ` IntermediateDataSetID` from
`SingleInputGate` and might replace `IntermediateDataSetID` by `InputGateID` in
`InputGateDeploymentDescriptor`.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
