GJL commented on a change in pull request #9663: [WIP][FLINK-12433][runtime]
Implement DefaultScheduler stub
URL: https://github.com/apache/flink/pull/9663#discussion_r324247575
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/executiongraph/Execution.java
##########
@@ -1211,7 +1231,10 @@ private boolean processFail(Throwable t, boolean
isCallback, Map<String, Accumul
return false;
}
- if (transitionState(current, FAILED, t)) {
+ if (!isCallback && !isLegacyScheduling()) {
Review comment:
The new `DefaultScheduler` needs to update internal state for every task
failure. This should happen via
`DefaultScheduler#updateTaskExecutionState(...)`. Normally the TaskManager will
invoke this method via a JobManager RPC. However, sometimes the task can be
failed for reasons that did not originate from the TaskManager. In such cases
the `DefaultScheduler` should still be informed about the Task failure – this
happens by invoking `vertex.getExecutionGraph().notifyTaskFailed(attemptId,
t)`. We do not want to inform the `DefaultScheduler` twice and the same task
failure, and the condition attempts to model this. I am having some doubts
about its correctness now. Maybe there are other means to filter double
invocations of `updateTaskExecutionState(...)`. For example, using the
attemptId.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
