[
https://issues.apache.org/jira/browse/SOLR-1972?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13539623#comment-13539623
]
Shawn Heisey commented on SOLR-1972:
------------------------------------
I've been trying to find evidence of the leak, and can't. I completely trust
that you guys do know what you're talking about, so this is probably just my
inexperience.
What I did find:
Before any additional patching, looking at jconsole, I have two metrics
threads. Looking at MBeans in jconsole, there are 290 metric scopes under
RequestHandlerBase with names from metrics-scope-0 to metrics-scope-289.
Taking Robert's patch and fixing it to create one static MetricsRegistry, the
additional objects under MBeans disappear, and the number of threads named
"metric-*" goes from 2 to 1. Through some additional debugs, I have files
containing stacktraces from every time RequestHandlerBase is called, so I know
it is still being called 290 times during startup.
Examining those stacktraces, I have determined that each of my 16 cores has 18
request handlers in common, with two additional server-wide handlers -
CollectionsHandler and CoreAdminHandler - for a total of 290. I'm attaching an
archive with all the stacktraces and a summary - the third line from each
stacktrace, sorted.
If we made the metrics package optional on a per-handler basis, I would drop
from 290 scopes to 48 - each of my 16 cores has four search handlers, but I
really only need full metrics in three of them.
> Need additional query stats in admin interface - median, 95th and 99th
> percentile
> ---------------------------------------------------------------------------------
>
> Key: SOLR-1972
> URL: https://issues.apache.org/jira/browse/SOLR-1972
> Project: Solr
> Issue Type: Improvement
> Components: web gui
> Affects Versions: 1.4
> Reporter: Shawn Heisey
> Assignee: Alan Woodward
> Priority: Minor
> Fix For: 4.1, 5.0
>
> Attachments: elyograg-1972-3.2.patch, elyograg-1972-3.2.patch,
> elyograg-1972-trunk.patch, elyograg-1972-trunk.patch, leak.patch,
> SOLR-1972-branch3x-url_pattern.patch, SOLR-1972-branch4x.patch,
> SOLR-1972-branch4x.patch, SOLR-1972_metrics.patch, SOLR-1972_metrics.patch,
> SOLR-1972_metrics.patch, SOLR-1972_metrics.patch, SOLR-1972_metrics.patch,
> SOLR-1972_metrics.patch, SOLR-1972_metrics.patch, SOLR-1972_metrics.patch,
> solr1972-metricsregistry-branch4x-failure.log, SOLR-1972.patch,
> SOLR-1972.patch, SOLR-1972.patch, SOLR-1972.patch, SOLR-1972-url_pattern.patch
>
>
> I would like to see more detailed query statistics from the admin GUI. This
> is what you can get now:
> requests : 809
> errors : 0
> timeouts : 0
> totalTime : 70053
> avgTimePerRequest : 86.59209
> avgRequestsPerSecond : 0.8148785
> I'd like to see more data on the time per request - median, 95th percentile,
> 99th percentile, and any other statistical function that makes sense to
> include. In my environment, the first bunch of queries after startup tend to
> take several seconds each. I find that the average value tends to be useless
> until it has several thousand queries under its belt and the caches are
> thoroughly warmed. The statistical functions I have mentioned would quickly
> eliminate the influence of those initial slow queries.
> The system will have to store individual data about each query. I don't know
> if this is something Solr does already. It would be nice to have a
> configurable count of how many of the most recent data points are kept, to
> control the amount of memory the feature uses. The default value could be
> something like 1024 or 4096.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
