Erick Erickson created SOLR-8301:
------------------------------------
Summary: We should distinguish between top-level queries and
sub-queries when reporting stats
Key: SOLR-8301
URL: https://issues.apache.org/jira/browse/SOLR-8301
Project: Solr
Issue Type: Improvement
Affects Versions: 5.4, Trunk
Reporter: Erick Erickson
It looks like we look at all query times when reporting plugins/stats. This
includes sub-queries in SolrCloud (i.e. distrib=false).
This doesn't really give a true picture of what the _user_ sees, i.e. it
doesn't really reflect "the laggard problem".
To diagnose a cluster's performance, it seems like it would be useful to
aggregate perf stats in two groups at least, the top level (i.e. distrib !=
false) and distrib=true.
Or perhaps even three, the aggregator times, the distrib=false grouping and
actually getting the stored data from the docs to return the client.
This is always available in the logs, but being able to monitor them with JMX
seems useful.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
