You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@ambari.apache.org by "Srimanth Gunturi (JIRA)" <ji...@apache.org> on 2016/01/16 01:12:39 UTC

[jira] [Commented] (AMBARI-14699) JMX heatmap API call takes over 1 minute to complete on 900 node cluster

    [ https://issues.apache.org/jira/browse/AMBARI-14699?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15102750#comment-15102750 ] 

Srimanth Gunturi commented on AMBARI-14699:
-------------------------------------------

There are some places where performance is not optimal
# ThreadPoolExecutor is not being used correctly where the core pool-size does not scale up to max-pool-size when needed. Due to this computations which can be done in parallel are not being done so efficiently. This issue has also been reported in AMBARI-14671.
# _JMXPropertyProvider_ in its attempt to get JMX values, first attempts to get the JMX port which takes 700ms on a 900 node cluster. When this is done for 900 DataNodes in 1 call, the delay is significant. Adding to this issue is the above problem of lesser number of threads.
# _AMSPropertyProvider_ for each of the 900 components repeatedly determines if the host and host-component are alive, when it can reuse the information from first call.

> JMX heatmap API call takes over 1 minute to complete on 900 node cluster
> ------------------------------------------------------------------------
>
>                 Key: AMBARI-14699
>                 URL: https://issues.apache.org/jira/browse/AMBARI-14699
>             Project: Ambari
>          Issue Type: Bug
>          Components: ambari-server
>    Affects Versions: 2.2.0
>            Reporter: Srimanth Gunturi
>            Assignee: Srimanth Gunturi
>             Fix For: 2.4.0
>
>
> On a 900 node cluster I went to the Dashboard heatmaps page for HDFS and the _DataNode Garbage Collection Time_ heatmap took over 1 minute to respond. URL call is shown below.
> {code}http://perf-a-1:8080/api/v1/clusters/perf/services/HDFS/components/DATANODE?fields=host_components/metrics/jvm/gcTimeMillis&format=null_padding{code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)