[
https://issues.apache.org/jira/browse/SOLR-10262?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15905360#comment-15905360
]
Andrzej Bialecki commented on SOLR-10262:
------------------------------------------
A few relevant links:
* [discussion
thread|https://groups.google.com/forum/#!msg/mechanical-sympathy/I4JfZQ1GYi8/ocuzIyC3N9EJ]
where Gil Tene (from Azul) explains why default reservoirs in Codahale Metrics
are suboptimal, to put it mildly...
* [How NOT to measure latency|
https://www.azul.com/files/HowNotToMeasureLatency_LLSummit_NYC_12Nov2013.pdf]
* [HrdHistogram
reservoir|https://bitbucket.org/marshallpierce/hdrhistogram-metrics-reservoir]
implementation. Notably it contains {{HdrHistogramReservoir}} which tracks
values since its creation, and {{HdrHistogramResetOnSnapshotReservoir}} which
resets its state to zero after each snapshot. There's no equivalent
exponentially-decayed reservoir, and we can't use the one that resets on
snapshot (because multiple clients may read from it at different intervals).
> Collect request latency metrics for histograms
> ----------------------------------------------
>
> Key: SOLR-10262
> URL: https://issues.apache.org/jira/browse/SOLR-10262
> Project: Solr
> Issue Type: New Feature
> Security Level: Public(Default Security Level. Issues are Public)
> Components: metrics
> Reporter: Otis Gospodnetic
> Assignee: Andrzej Bialecki
>
> Since [~ab] is on a role with metrics...
> There is no way to accurately compute request latency percentiles from
> metrics exposed by Solr today. We should consider making that possible. c.f.
> https://github.com/HdrHistogram/HdrHistogram
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
