Matthias Pohl created FLINK-33500:
-------------------------------------
Summary: Run storing the JobGraph an asynchronous operation
Key: FLINK-33500
URL: https://issues.apache.org/jira/browse/FLINK-33500
Project: Flink
Issue Type: Improvement
Components: Runtime / Coordination
Affects Versions: 1.17.1, 1.18.0, 1.19.0
Reporter: Matthias Pohl
Currently, submitting a job starts with storing the JobGraph (in HA setups) in
the {{{}JobGraphStore{}}}. This includes writing the file to S3 (or some other
remote file system). The job submission is done in the {{{}Dispatcher{}}}'s
main thread. If writing the {{JobGraph}} is slow, it would block any other
operation on the {{{}Dispatcher{}}}. See
[Dispatcher#persistAndRunJob|https://github.com/apache/flink/blob/52cbeb90f32ca36c59590df1daa6748995c9b7f8/flink-runtime/src/main/java/org/apache/flink/runtime/dispatcher/Dispatcher.java#L645]
as code reference.
This Jira issue is about moving the job submission into the {{ioExecutor}} as
an asynchronous call.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
