Github user sachingoel0101 commented on a diff in the pull request:
https://github.com/apache/flink/pull/979#discussion_r36188838
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/client/JobClientActor.java
---
@@ -59,11 +87,50 @@ public JobClientActor(
this.leaderSessionID =
Preconditions.checkNotNull(leaderSessionID, "The leader session ID option must
not be null.");
this.sysoutUpdates = sysoutUpdates;
+ // set this to -1 to indicate the job hasn't been created yet.
+ this.currentJobCreatedAt = -1;
}
@Override
protected void handleMessage(Object message) {
+ // ======= Job status messages on regular intervals
==============
+ if(message instanceof JobManagerMessages.CurrentJobStatus){
+ JobStatus statusReport =
((JobManagerMessages.CurrentJobStatus) message).status();
+ long timeDiff;
+ switch(statusReport){
+ case RUNNING:
+ // Vincent, we happy?
+ this.currentJobCreatedAt = -1;
+ break;
+ case FINISHED:
+ // Yeah! We happy!
+ this.currentJobCreatedAt = -1;
+ break;
+ case CREATED:
+ // we're still at Job CREATED. Let's
see if we're over the limit.
+ timeDiff = (System.currentTimeMillis()
- this.currentJobCreatedAt);
+ if(timeDiff >
JOB_CLIENT_JOB_STATUS_TIMEOUT){
+ failWithTimeout(timeDiff);
+ } // otherwise just wait a bit longer.
+ break;
+ case RESTARTING:
+ if(this.currentJobCreatedAt == -1){
+ // we effectively have
re-created the job
+ this.currentJobCreatedAt =
System.currentTimeMillis();
+ } else{
+ // it was already at either
CREATED or RESTARTING. See if we're over the limit.
+ timeDiff =
(System.currentTimeMillis() - this.currentJobCreatedAt);
+ if(timeDiff >
JOB_CLIENT_JOB_STATUS_TIMEOUT){
+
failWithTimeout(timeDiff);
+ } // otherwise let's wait a bit
more.
+ }
+ break;
+ default:
+ // well, canceled or failed. We'll find
out with an exact message in a while.
--- End diff --
Okay, that makes sense. Here's what I propose. If the job doesn't change
status from the cancel(ed/ing) or fail(ed/ing) states for a given timeout
period, we again send a failure message to the `JobClient`.
The reason I kept this as is was that if the job has been canceled or
failed, then the `JobClient` should know the reason for failure, which can only
be sent by the `JobManager`. It still seems logical to me though.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
