[
https://issues.apache.org/jira/browse/SOLR-17799?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17987221#comment-17987221
]
Matthew Biscocho commented on SOLR-17799:
-----------------------------------------
Might need to keep them. If more "minor" merges are happening than "major", it
may skew the timers quantiles. There is no way that I know of in a metrics
store or promQL that can make that distinction. That distinction probably has
to stay in Solr. Maybe if you could do a prometheus histogram that counted
merges differed by a # of document buckets and somehow aggregated the timer and
them together. I'm just throwing ideas though, not sure if a query like that is
actually possible to correlate the 2 but maybe a metric like that is still
useful. Something like this shows 5 merges happened with documents < 100k and 6
merges happened with documents < 500k :
docs_per_merge\{le=100,000} 5
docs_per_merge\{le=500,000} 6
> Revamp segment merge metrics
> ----------------------------
>
> Key: SOLR-17799
> URL: https://issues.apache.org/jira/browse/SOLR-17799
> Project: Solr
> Issue Type: Improvement
> Reporter: Kevin Liang
> Priority: Minor
>
> Based on current
> [documentation|https://solr.apache.org/guide/solr/latest/deployment-guide/metrics-reporting.html#index-merge-metrics]
> for segment merge metrics, there are two major groups of metrics emitted
> (for currently actively running merges and for merges that have completed).
> The actively running ones are gauges that can go up or down, while the ones
> for completed are counters/timers.
> To make these metrics more consistent and aggregate-able, propose emitting
> all metrics for merges as running counters for # of merges, # of segments,
> and # of documents with a metric for each tracking started/completed (thus
> active merges can be calculated via the current diff between the two).
> Relates to original feature from
> https://issues.apache.org/jira/browse/SOLR-9854
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
