[ 
https://issues.apache.org/jira/browse/SOLR-6752?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14218077#comment-14218077
 ] 

Mike Drob commented on SOLR-6752:
---------------------------------

bq. I'm a little confused by the title - isn't this simply exposing hdfds block 
cache metrics via JMX? Why anything specific about allocate/lost? I see all 
sorts of stats in the getStatistics call.
The buffer allocatione/lost metrics are currently not exposed at all. When 
moving everything from {{metricsRecord.setMetric}} to {{stats.add}} in the 
metrics method, these lines are completely new, instead of being converted over.
{noformat}
+      stats.add("buffercache.allocations", 
getPerSecond(shardBuffercacheAllocate.getAndSet(0), seconds));
+      stats.add("buffercache.lost", 
getPerSecond(shardBuffercacheLost.getAndSet(0), seconds));
{noformat}

bq. We also probably want to try and align some of the stat key names with 
other cache objects in Solr: the query cache, filter cache, etc.
This makes sense, I'll look into it.

bq. Also, I don't believe these will get registered with the jmx server. I 
think only the top level class for a plugin is by default - eg the 
HdfsDirectoryFactory itself.
Not sure I understand this. Where should I look to make sure this is getting 
registered?



> Buffer Cache allocate/lost should be exposed through JMX
> --------------------------------------------------------
>
>                 Key: SOLR-6752
>                 URL: https://issues.apache.org/jira/browse/SOLR-6752
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Mike Drob
>            Assignee: Mark Miller
>              Labels: metrics
>         Attachments: SOLR-6752.patch
>
>
> Currently, {{o.a.s.store.blockcache.Metrics}} has fields for tracking buffer 
> allocations and losses, but they are never updated nor exposed to a receiving 
> metrics system. We should do both. 



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to