[ 
https://issues.apache.org/jira/browse/STORM-533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14200252#comment-14200252
 ] 

ASF GitHub Bot commented on STORM-533:
--------------------------------------

Github user revans2 commented on the pull request:

    https://github.com/apache/storm/pull/302#issuecomment-61989690
  
    The metrics system is very generic, and not that complex.  Essentially it 
sets up a timer that will periodically call getValueAndReset on an instance of 
IMetric.  These values can be anything and are sent to an instance of 
IMetricsConsumer that is residing in a bolt.  It is up to the IMetricsConsumer 
to decide what to do with Object the the IMetric created.
    
    I agree that having an API closer to codahale would be good, but that is a 
much bigger change.  I would like to see that in a separate JIRA/pull request.  
The big difference between the two approaches is that the storm metrics 
associate the value with an individual bolt or spout instance.  codahale and 
most other metrics systems I have seen, associate the metrics with an arbitrary 
name.  We would need a way to bridge that gap in a clean/efficient way.  I 
would also like to see a lot of metrics added into the daemon processes.  
Nimbus, Supervisor, and DRPC all need good monitoring.


> Add metrics collection for IConnection
> --------------------------------------
>
>                 Key: STORM-533
>                 URL: https://issues.apache.org/jira/browse/STORM-533
>             Project: Apache Storm
>          Issue Type: Improvement
>            Reporter: Robert Joseph Evans
>            Assignee: Robert Joseph Evans
>
> It would really be great to get some metrics from an IConnection that are 
> then sent to the metrics consumer. 
> We have seen issues in the past where a fire wall rule is mis-configured and 
> one host is unable to talk to another host.  If we had some metrics about how 
> many reconnection attempts are being made by the client to a given host we 
> could easily diagnose this.
> There are other metrics that would be nice to know too, like how many 
> bytes/tuples are being sent between different hosts. 



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

Reply via email to