[
https://issues.apache.org/jira/browse/GOBBLIN-2189?focusedWorklogId=955334&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-955334
]
ASF GitHub Bot logged work on GOBBLIN-2189:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 04/Feb/25 04:50
Start Date: 04/Feb/25 04:50
Worklog Time Spent: 10m
Work Description: abhishekmjain commented on code in PR #4092:
URL: https://github.com/apache/gobblin/pull/4092#discussion_r1940501751
##########
gobblin-temporal/src/main/java/org/apache/gobblin/temporal/yarn/DynamicScalingYarnService.java:
##########
@@ -232,7 +237,12 @@ private synchronized void
handleContainerExitedWithOOM(ContainerId completedCont
// Request a replacement container
int currContainerMemoryMbs =
workerProfile.getConfig().getInt(GobblinYarnConfigurationKeys.CONTAINER_MEMORY_MBS_KEY);
- int newContainerMemoryMbs = currContainerMemoryMbs * 2; //TODO: make it
configurable or auto-tunable
+ int newContainerMemoryMbs = currContainerMemoryMbs *
DEFAULT_REPLACEMENT_CONTAINER_MEMORY_MULTIPLIER;
+ if (newContainerMemoryMbs > MAX_REPLACEMENT_CONTAINER_MEMORY_MBS) {
+ log.warn("Expected replacement container memory exceeds the maximum
allowed memory {}. Not requesting a replacement container.",
+ MAX_REPLACEMENT_CONTAINER_MEMORY_MBS);
+ return;
Review Comment:
We have two options here:
1. to create a replacement container with same memory as the old container
2. return without creating any container.
I think 2 here would help us in failing the flow fast, whereas 1 would help
resolve any intermittent OOM issues (which is unlikely). We have opted for 2
here, but may need to update this in future based on the observations.
Issue Time Tracking
-------------------
Worklog Id: (was: 955334)
Time Spent: 1h 40m (was: 1.5h)
> Implement ContainerCompletion callback in DynamicScalingYarnService
> -------------------------------------------------------------------
>
> Key: GOBBLIN-2189
> URL: https://issues.apache.org/jira/browse/GOBBLIN-2189
> Project: Apache Gobblin
> Issue Type: Improvement
> Components: gobblin-core
> Reporter: Vivek Rai
> Assignee: Abhishek Tiwari
> Priority: Major
> Time Spent: 1h 40m
> Remaining Estimate: 0h
>
> DynamicScalingYarnService currently doesn't handle scaling down containers
> and neither does anything if container is killed abruptly or goes OOM. So to
> handle this scenario containerCompletion callback should be implemented to
> launch the replacement containers and also scaling down handling should be
> done.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
