[
https://issues.apache.org/jira/browse/MAPREDUCE-1779?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12868269#action_12868269
]
Arun C Murthy commented on MAPREDUCE-1779:
------------------------------------------
bq. It is also needed to monitor the health of the JT.
This is already available in JVM Metrics.
bq. Hive actually throttles new job-submissions if the heap memory on the JT
exceeds a certain threshold.
Again, I'd like to re-iterate that this 'feature' causes *serious* performance
issues on the JobTracker - JNI calls are *very* expensive, a rouge client for
this feature can easily cause *severe* harm to the JobTracker and hence the
entire cluster.
Hive can use JVM Metrics for the same functionality given that this is already
available in JVM Metrics. Thus, I'm -1 on this feature.
> Should we provide a way to know JobTracker's memory info from client?
> ---------------------------------------------------------------------
>
> Key: MAPREDUCE-1779
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-1779
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: client, jobtracker
> Affects Versions: 0.21.0
> Reporter: Amareshwari Sriramadasu
> Priority: Blocker
> Fix For: 0.22.0
>
>
> In HADOOP-4435, in branch 0.20, getClusterStatus() method returns
> JobTracker's used memory and total memory.
> But these details are missed in new api (through MAPREDUCE-777).
> If these details are needed only for web UI, I don't think they are needed
> for client.
> So, should we provide a way to know JobTracker's memory info from client?
> If yes, an api should be added in org.apache.hadoop.mapreduce.Cluster for the
> same.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
