[
https://issues.apache.org/jira/browse/GOBBLIN-1728?focusedWorklogId=818551&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-818551
]
ASF GitHub Bot logged work on GOBBLIN-1728:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 19/Oct/22 17:43
Start Date: 19/Oct/22 17:43
Worklog Time Spent: 10m
Work Description: homatthew commented on code in PR #3586:
URL: https://github.com/apache/gobblin/pull/3586#discussion_r999766182
##########
gobblin-yarn/src/main/java/org/apache/gobblin/yarn/YarnService.java:
##########
@@ -789,11 +815,17 @@ private ImmutableMap.Builder<String, String>
buildContainerStatusEventMetadata(C
/**
* Get the number of matching container requests for the specified resource
memory and cores.
+ * Due to YARN-1902, this API is not 100% accurate. However, {@link
AMRMClientCallbackHandler#onContainersAllocated(List)}
+ * contains logic for best effort clean up of duplicate requests, and in
practice is pretty accurate.
*/
private int getMatchingRequestsCount(Resource resource) {
int priorityNum = resourcePriorityMap.getOrDefault(resource.toString(), 0);
Priority priority = Priority.newInstance(priorityNum);
- return getAmrmClientAsync().getMatchingRequests(priority,
ResourceRequest.ANY, resource).size();
+ List<? extends Collection> outstandingRequests =
getAmrmClientAsync().getMatchingRequests(priority, ResourceRequest.ANY,
resource);
Review Comment:
The other logical change is that this API call `getMatchingRequests` returns
a `List<? extends Collection>` and we were calculating the number of
outstanding requests incorrectly.
If we just return the size of this, we will always return a small value
because the API returns a list of collections, where each collection is the
requests partitioned by the node ID. So
The pending requests count should be the sum of all outstanding requests.
Issue Time Tracking
-------------------
Worklog Id: (was: 818551)
Time Spent: 1.5h (was: 1h 20m)
> 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: 1.5h
> 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)
