[
https://issues.apache.org/jira/browse/GOBBLIN-1728?focusedWorklogId=818545&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-818545
]
ASF GitHub Bot logged work on GOBBLIN-1728:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 19/Oct/22 17:36
Start Date: 19/Oct/22 17:36
Worklog Time Spent: 10m
Work Description: homatthew commented on code in PR #3586:
URL: https://github.com/apache/gobblin/pull/3586#discussion_r999759167
##########
gobblin-yarn/src/main/java/org/apache/gobblin/yarn/YarnService.java:
##########
@@ -845,8 +877,6 @@ public void onContainersAllocated(List<Container>
containers) {
instanceName = null;
}
}
- allocatedContainerCountMap.put(containerHelixTag,
Review Comment:
Putting this block inside of the synchronized block works is kind of
confusing. I am not entirely convinced it does what we want it to. (Does
synchronized sync on the `AmrmClient` or does it sync on the `YarnService`. I
think it would be the former which is incorrect.
With this being a relatively cold part of the code (once every 90 seconds?),
we really don't need any fine-grained locking. An atomic integer would be
sufficient or even a global lock for all instances of the class. (YarnService
is a single thread and then there is a callback for each container allocated.
Container allocation / deallocation doesn't happen that much especially after
the pipeline is up for a few hours.
Issue Time Tracking
-------------------
Worklog Id: (was: 818545)
Time Spent: 40m (was: 0.5h)
> Yarn Service requests too many containers due to improper calculation
> ---------------------------------------------------------------------
>
> Key: GOBBLIN-1728
> URL: https://issues.apache.org/jira/browse/GOBBLIN-1728
> Project: Apache Gobblin
> Issue Type: New Feature
> Reporter: Matthew Ho
> Priority: Major
> Time Spent: 40m
> Remaining Estimate: 0h
>
> Yarn Service is responsible for calculating the number of instances based on
> the helix tasks. Yarn service tracks the number of instances by asking Yarn
> for the number of resource requests and the number of allocated containers.
>
> It uses this count to determine if it should ask for more containers or
> shrink the number of containers. This calculation is currently done
> improperly and we continue to request containers when we have enough
> requested.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
