xintongsong commented on a change in pull request #9105:
[FLINK-13241][Yarn/Mesos] Fix Yarn/MesosResourceManager setting managed memory
size into wrong configuration instance.
URL: https://github.com/apache/flink/pull/9105#discussion_r307812100
##########
File path:
flink-yarn/src/main/java/org/apache/flink/yarn/YarnResourceManager.java
##########
@@ -184,7 +183,9 @@ public YarnResourceManager(
this.defaultCpus =
flinkConfig.getInteger(YarnConfigOptions.VCORES, numberOfTaskSlots);
this.resource =
Resource.newInstance(defaultTaskManagerMemoryMB, defaultCpus);
- this.slotsPerWorker =
updateTaskManagerConfigAndCreateWorkerSlotProfiles(flinkConfig,
defaultTaskManagerMemoryMB, numberOfTaskSlots);
+ this.defaultTaskManagerParameters =
ContaineredTaskManagerParameters.create(flinkConfig,
defaultTaskManagerMemoryMB, numberOfTaskSlots);
Review comment:
Yes, the `createTaskExecutorLaunchContext ` recreates a
`ContaineredTaskManagerParameters`, but with the actual container memory
allocated by Yarn, which could be slightly larger than requested. Reusing the
same `ContaineredTaskManagerParameters` instance prevents task managers from
making use of the extra allocated memory, which could be a user aware behavior
change compare to the current behavior. I think maybe we should not introduce
such behavior change in this version, since we are going to change resource
configuration and management anyway in the next version.
----------------------------------------------------------------
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
