[jira] [Created] (FLINK-12890) Add partition lifecycle related Shuffle API

["Andrey Zagrebin (JIRA)" <ji...@apache.org>]

Andrey Zagrebin created FLINK-12890:
---------------------------------------

             Summary: Add partition lifecycle related Shuffle API
                 Key: FLINK-12890
                 URL: https://issues.apache.org/jira/browse/FLINK-12890
             Project: Flink
          Issue Type: Sub-task
          Components: Runtime / Coordination
            Reporter: Andrey Zagrebin
            Assignee: Andrey Zagrebin
             Fix For: 1.9.0


At the moment we have ShuffleEnvironment.releasePartitions which is used to release locally occupied resources of partition. JM can also use it by calling TaskExecutorGateway.releasePartitions.

To support lifecycle management of partitions (FLINK-12069, relevant mostly for batch and blocking partitions), we need to extend Shuffle API:
 * ShuffleDescriptor.hasLocalResources() indicates that this partition occupies local resources on TM and requires TM running to consume the produced data (e.g. true for default NettyShuffleEnviroment and false for externally stored partitions). If a partition needs external lifecycle management and is not released after the first consumption is done (ResultPartitionDeploymentDescriptor.isReleasedOnConsumption()), then RM/JM should keep TMs, which produce these partitions, running until partition still needs to be consumed. The connection to these TMs should also to be kept to issue the RPC call TaskExecutorGateway.releasePartitions once partition is not needed any more.

 * ShuffleMaster.removePartitionExternally(): JM should call this whenever the partition doe not need to be consumed any more. This call releases partition resources possibly occupied externally outside of TM and should not depend on ShuffleDescriptor.hasLocalResources.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)