[
https://issues.apache.org/jira/browse/YARN-11238?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17607677#comment-17607677
]
ASF GitHub Bot commented on YARN-11238:
---------------------------------------
slfan1989 commented on code in PR #4904:
URL: https://github.com/apache/hadoop/pull/4904#discussion_r976348794
##########
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-router/src/test/java/org/apache/hadoop/yarn/server/router/clientrm/TestFederationClientInterceptor.java:
##########
@@ -582,9 +583,9 @@ public void testGetClusterMetricsRequest() throws Exception
{
ClientMethod remoteMethod = new ClientMethod("getClusterMetrics",
new Class[] {GetClusterMetricsRequest.class},
new Object[] {GetClusterMetricsRequest.newInstance()});
- Map<SubClusterId, GetClusterMetricsResponse> clusterMetrics = interceptor.
- invokeConcurrent(new ArrayList<>(), remoteMethod,
GetClusterMetricsResponse.class);
- Assert.assertTrue(clusterMetrics.isEmpty());
+ Collection<GetClusterMetricsResponse> clusterMetrics = interceptor.
+ invokeConcurrent(remoteMethod, GetClusterMetricsResponse.class);
+ Assert.assertTrue(!clusterMetrics.isEmpty());
Review Comment:
`invokeConcurrent` This function originally contains 3 parameters
1. ArrayList<SubClusterId> clusterIds
2. ClientMethod request
3. Class<R> clazz
The first parameter is the SubClusters currently active.
In order to get the first parameter, all calling functions need to call the
following function
`Map<SubClusterId, SubClusterInfo> subclusters =
federationFacade.getSubClusters(true);`
I think we can put the fetching of the 1st parameter inside
`invokeConcurrent`, which is more reasonable.
This unit test tests a situation where there is no Subcluster available, so
an empty result is returned.
```
Map<SubClusterId, GetClusterMetricsResponse> clusterMetrics = interceptor.
invokeConcurrent(new ArrayList<>(), remoteMethod,
GetClusterMetricsResponse.class);
```
I removed the 1st parameter, it is a little difficult to simulate the
situation where all Subclusters are unavailable.
I try to ensure that the original test logic can be completed, and I will
refactor part of the code.
##########
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-router/src/test/java/org/apache/hadoop/yarn/server/router/clientrm/TestFederationClientInterceptor.java:
##########
@@ -582,9 +583,9 @@ public void testGetClusterMetricsRequest() throws Exception
{
ClientMethod remoteMethod = new ClientMethod("getClusterMetrics",
new Class[] {GetClusterMetricsRequest.class},
new Object[] {GetClusterMetricsRequest.newInstance()});
- Map<SubClusterId, GetClusterMetricsResponse> clusterMetrics = interceptor.
- invokeConcurrent(new ArrayList<>(), remoteMethod,
GetClusterMetricsResponse.class);
- Assert.assertTrue(clusterMetrics.isEmpty());
+ Collection<GetClusterMetricsResponse> clusterMetrics = interceptor.
+ invokeConcurrent(remoteMethod, GetClusterMetricsResponse.class);
+ Assert.assertTrue(!clusterMetrics.isEmpty());
Review Comment:
I will fix it.
> Optimizing FederationClientInterceptor Call with Parallelism
> ------------------------------------------------------------
>
> Key: YARN-11238
> URL: https://issues.apache.org/jira/browse/YARN-11238
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: federation
> Affects Versions: 3.4.0, 3.3.4
> Reporter: fanshilun
> Assignee: fanshilun
> Priority: Major
> Labels: pull-request-available
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
