[
https://issues.apache.org/jira/browse/FLINK-24824?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17453892#comment-17453892
]
David Morávek commented on FLINK-24824:
---------------------------------------
{code}
/**
* The job has been received by the Dispatcher, and is waiting for the job
manager to receive
* leadership and to be created.
*/
INITIALIZING(TerminalState.NON_TERMINAL),
/** Job is newly created, no task has started to run. */
CREATED(TerminalState.NON_TERMINAL),
{code}
[~chesnay] Makes sense to me. We should stay as close to the original contract
as possible. These javadocs imply, that we should switch to CREATED state right
after we have an active JobMaster for the job.
> AdaptiveScheduler#getJobStatus never returns CREATED
> ----------------------------------------------------
>
> Key: FLINK-24824
> URL: https://issues.apache.org/jira/browse/FLINK-24824
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
> Reporter: Chesnay Schepler
> Assignee: Chesnay Schepler
> Priority: Major
> Fix For: 1.15.0
>
>
> None of the AdaptiveScheduler states return {{JobStatus#CREATED}} when
> {{requestJobStatus()}} is called. This violates the job state machine, and
> makes it a bit difficult to setup job state timestamps in the
> {{AdaptiveScheduler}} (FLINK-24775).
> We are in the {{INITIALIZING}} state from {{Created}} ->
> {{WaitingForResources}} -> {{CreatingExecutionGraph}}, and then switch
> straight to RUNNING in {{Executing}}.
> It is tricky to retain the same semantics for the {{Default}} and
> {{AdaptiveScheduler}}, but I think it would be fine to return {{CREATED}}
> once we reached {{WaitingForResources}} because from the users-perspective it
> behaves similarly.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
