You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-issues@hadoop.apache.org by "Franklin Ta (JIRA)" <ji...@apache.org> on 2010/07/22 07:45:49 UTC
[jira] Updated: (HADOOP-6874) JvmMetrics can't distinguish between
jvms with same processName
[ https://issues.apache.org/jira/browse/HADOOP-6874?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Franklin Ta updated HADOOP-6874:
--------------------------------
Description:
JvmMetrics has three tags: hostName, processName, and sessionId. For processes such as tasktracker/jobtracker/namenode/datanode which there is only one of on each host, these tags are fine. But for process names such as "MAP" and "REDUCE", since there might be multiple jvms running map/reduce tasks, we might end up with multiple set of metrics which all have the same tags, and no way to find out which jvm they actually correspond to. (In addition, since there is jvm reuse, those process names might not correspond to the actual task being ran)
A quick fix is to change this line in Child.java
JvmMetrics.init(task.getPhase().toString(), job.getSessionId());
to this
JvmMetrics.init(jvmId.toString(), job.getSessionId());
so that we are using the jvm id for the process name instead.
was:
JvmMetrics has three tags: hostName, processName, and sessionId. For processes such tasktracker/jobtracker/namenode/datanode which there is only one of on each host, these tags are fine. But for process names such as "MAP" and "REDUCE", since there might be multiple jvms running map/reduce tasks, we might end up with multiple set of metrics which all have the same tags, and no way to find out which jvm they actually correspond to. (In addition, since there is jvm reuse, those process names might not correspond to the actual task being ran)
A quick fix is to change this line in Child.java
JvmMetrics.init(task.getPhase().toString(), job.getSessionId());
to this
JvmMetrics.init(jvmId.toString(), job.getSessionId());
so that we are using the jvm id for the process name instead.
> JvmMetrics can't distinguish between jvms with same processName
> ---------------------------------------------------------------
>
> Key: HADOOP-6874
> URL: https://issues.apache.org/jira/browse/HADOOP-6874
> Project: Hadoop Common
> Issue Type: Bug
> Components: metrics
> Affects Versions: 0.20.2
> Reporter: Franklin Ta
> Priority: Minor
>
> JvmMetrics has three tags: hostName, processName, and sessionId. For processes such as tasktracker/jobtracker/namenode/datanode which there is only one of on each host, these tags are fine. But for process names such as "MAP" and "REDUCE", since there might be multiple jvms running map/reduce tasks, we might end up with multiple set of metrics which all have the same tags, and no way to find out which jvm they actually correspond to. (In addition, since there is jvm reuse, those process names might not correspond to the actual task being ran)
> A quick fix is to change this line in Child.java
> JvmMetrics.init(task.getPhase().toString(), job.getSessionId());
> to this
> JvmMetrics.init(jvmId.toString(), job.getSessionId());
> so that we are using the jvm id for the process name instead.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.