[ 
https://issues.apache.org/jira/browse/YARN-3628?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14542988#comment-14542988
 ] 

zhihai xu commented on YARN-3628:
---------------------------------

Sorry for the confusion. It is jmx.
The following is empty container metrics shown from the jmx
{code}
{
    "name" : 
"Hadoop:service=NodeManager,name=ContainerResource_container_1431465632950_0001_01_000005",
    "modelerType" : "ContainerResource_container_1431465632950_0001_01_000005"
  }
{
    "name" : 
"Hadoop:service=NodeManager,name=ContainerResource_container_1431465632950_0001_01_000002",
    "modelerType" : "ContainerResource_container_1431465632950_0001_01_000002"
  }
{code}

Normally the full container metrics will show like this:
{code}
{
    "name" : 
"Hadoop:service=NodeManager,name=ContainerResource_container_1431465632950_0001_01_000024",
    "modelerType" : "ContainerResource_container_1431465632950_0001_01_000024",
    "tag.ContainerResource" : "container_1431465632950_0001_01_000024",
    "tag.Context" : "container",
    "tag.ContainerPid" : "18748",
    "tag.Hostname" : "XXXXXXXX",
    "PMemUsageNumUsage" : 3,
    "PMemUsageAvgMBs" : 71.66666666666667,
    "PMemUsageStdevMBs" : 47.42713709821976,
    "PMemUsageIMinMBs" : 19.0,
    "PMemUsageIMaxMBs" : 111.0,
    "PMemUsageMinMBs" : 19.0,
    "PMemUsageMaxMBs" : 111.0,
    "pMemLimit" : 1024,
    "vMemLimit" : 2150,
    "vCoreLimit" : 1
  }
{code}

> ContainerMetrics should support always-flush mode.
> --------------------------------------------------
>
>                 Key: YARN-3628
>                 URL: https://issues.apache.org/jira/browse/YARN-3628
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: zhihai xu
>            Assignee: zhihai xu
>         Attachments: YARN-3628.000.patch
>
>
> ContainerMetrics should support always-flush mode.
> It will be good to set ContainerMetrics as always-flush mode if 
> yarn.nodemanager.container-metrics.period-ms is configured as 0.
> Currently both 0 and -1 mean flush on completion.
> Also the current default value for 
> yarn.nodemanager.container-metrics.period-ms is -1 and the default value  for 
> yarn.nodemanager.container-metrics.enable is true. So the empty content is 
> shown for the active container metrics until it is finished.
> The default value for yarn.nodemanager.container-metrics.period-ms should not 
> be -1.
> flushOnPeriod is always false if flushPeriodMs is -1,
> the content will only be shown when the container is finished.
> {code}
>     if (finished || flushOnPeriod) {
>       registry.snapshot(collector.addRecord(registry.info()), all);
>     }
> {code}



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

Reply via email to