xintongsong commented on a change in pull request #9242: [FLINK-13408][runtime]
Schedule StandaloneResourceManager.setFailUnfulfillableRequest whenever the
leadership is acquired
URL: https://github.com/apache/flink/pull/9242#discussion_r307991610
##########
File path:
flink-mesos/src/main/java/org/apache/flink/mesos/runtime/clusterframework/MesosResourceManager.java
##########
@@ -201,7 +201,6 @@ public MesosResourceManager(
final ContaineredTaskManagerParameters
containeredTaskManagerParameters =
taskManagerParameters.containeredParameters();
this.slotsPerWorker =
updateTaskManagerConfigAndCreateWorkerSlotProfiles(
flinkConfig,
containeredTaskManagerParameters.taskManagerTotalMemoryMB(),
containeredTaskManagerParameters.numSlots());
- setFailUnfulfillableRequest(true);
Review comment:
Yes, unfulfillable requests should have failed right away for Yarn and
Mesos.
We have `SlotManagerFailUnfulfillableTest` that tests this behavior at the
scope of slot manager, but no tests at the scope of resource manager for Yarn
and Mesos.
I've update the PR changing the default value of
`SlotManager#failUnfulfillableRequest` to true and set it to false on start of
`StandaloneResourceManager`. So the Yarn / Mesos resource manager do not need
to do anything special regarding failing unfulfillable requests. I don't think
we need more test cases other than `SlotManagerFailUnfulfillableTest` to test
this behavior for Yarn / Mesos.
----------------------------------------------------------------
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
