[ https://issues.apache.org/jira/browse/HADOOP-7324?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13055616#comment-13055616 ]
Luke Lu commented on HADOOP-7324: --------------------------------- Thanks for the nice patch, Priyo! Here is a preliminary review: # Like I mentioned in our email exchange, I'm hesitant to make the impl classes public and use them directly in the sink. It breaks abstraction that makes certain later optimizations impossible. A better approach would be using the builtin visitor interface. That said, it might be OK in 0.20.20x series, as they're mostly in maintenance mode. # In the #init, please log the exception instead of printStackTrace. # The #getTag method is only used to get the context of the record, which is unnecessary, as the record has a #context method. # The #parseSocket method is unnecessary, please reuse metrics2.util.Servers#parse. # The sparse/dense logic in #putMetrics can be improved a bit: the cache is looked up twice in dense mode. You can have #update method return the record once in dense mode. # Ganglia treats a metric as counter if slope is positive and gauge for anything else. Having to specify slope explicitly in the conf for every counter is a tedious chore we can avoid. This is a deficiency in the existing metrics1 implementation as well. > Ganglia plugins for metrics v2 > ------------------------------ > > Key: HADOOP-7324 > URL: https://issues.apache.org/jira/browse/HADOOP-7324 > Project: Hadoop Common > Issue Type: Bug > Components: metrics > Affects Versions: 0.20.203.0, 0.23.0 > Reporter: Luke Lu > Priority: Blocker > Labels: regression > Fix For: 0.23.0 > > Attachments: HADOOP-7324.patch > > > Although, all metrics in metrics v2 are exposed via the standard JMX > mechanisms, most users are using Ganglia to collect metrics. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira