[jira] [Created] (FLINK-20095) SchedulerBase.stopWithSavepoint schedules operation after ExecutionGraph terminates

["Till Rohrmann (Jira)" <ji...@apache.org>]

Till Rohrmann created FLINK-20095:
-------------------------------------

             Summary: SchedulerBase.stopWithSavepoint schedules operation after ExecutionGraph terminates
                 Key: FLINK-20095
                 URL: https://issues.apache.org/jira/browse/FLINK-20095
             Project: Flink
          Issue Type: Improvement
          Components: Runtime / Coordination
    Affects Versions: 1.12.0
            Reporter: Till Rohrmann
             Fix For: 1.13.0


While looking into FLINK-20065, we realized that we cannot implement the change described in FLINK-20094 because the {{SchedulerBase}} schedules an asynchronous operation after the termination of the {{ExecutionGraph}}. This asynchronous operation needs to complete before the savepoint path is returned.

In order to solve FLINK-20094, we either need to keep track of these operations and suspend the {{SchedulerBase}} termination until all operations have completed or we might be to return the savepoint path without depending it on the completion of the {{handleAsync}} here: https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SchedulerBase.java#L921.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)