Andrey Zagrebin created FLINK-13371:
---------------------------------------
Summary: Release partitions in JM of producer gets restarted
Key: FLINK-13371
URL: https://issues.apache.org/jira/browse/FLINK-13371
Project: Flink
Issue Type: Bug
Components: Runtime / Coordination, Runtime / Network
Affects Versions: 1.9.0
Reporter: Andrey Zagrebin
As discussed in FLINK-13245, there can be a case that producer does not even
detect any consumption attempt if consumer fails before the connection is
established. It means we cannot fully rely on shuffle service for the release
on consumption in case of consumer failure. When producer restarts it will leak
partitions from the previous attempt. Previously we had an explicit release
call for this case in Execution.cancel/suspend. Basically JM has to explicitly
release all partitions produced by the previous task execution attempt in case
of producer restart, including `released on consumption` partitions. For this
change, we might need to track all partitions in PartitionTrackerImpl.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
