Ufuk Celebi created FLINK-10743:
-----------------------------------
Summary: Use 0 processExitCode for ApplicationStatus.CANCELED
Key: FLINK-10743
URL: https://issues.apache.org/jira/browse/FLINK-10743
Project: Flink
Issue Type: Bug
Components: Cluster Management, Kubernetes, Mesos, YARN
Affects Versions: 1.6.3
Reporter: Ufuk Celebi
Assignee: Ufuk Celebi
{{org.apache.flink.runtime.clusterframework.ApplicationStatus}} is used to map
{{org.apache.flink.runtime.jobgraph.JobStatus}} to a process exit code.
We currently map {{ApplicationStatus.CANCELED}} to a non-zero exit code
({{1444}}). Since cancellation is a user-triggered operation I would consider
this to be a successful exit and map it to exit code {{0}}.
Our current behavior results in applications running via the
{{StandaloneJobClusterEntryPoint}} and Kubernetes pods as documented in
[flink-container|https://github.com/apache/flink/tree/master/flink-container/kubernetes]
to be immediately restarted when cancelled. This only leaves the option of
killing the respective job cluster master container.
The {{ApplicationStatus}} is also used in the YARN and Mesos clients, but I'm
not familiar with that part of the code base and can't asses how changing the
exit code would affect these clients. A quick usage scan for
{{ApplicationStatus.CANCELED}} did not surface any problematic usages though.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
