Ok, my mistake -- aparently this was all heavily changed in 9.0 and I didn't notice until you asked about it...
https://issues.apache.org/jira/browse/SOLR-14401 https://solr.apache.org/guide/solr/9_1/deployment-guide/performance-statistics-reference.html So going back to my previous comment... > Metrics like "QUERY./select.distrib.requestTimes" tell you the stats on > handling a "distributed" request -- which is when a core is responsible to > sending out "per-shard" requests and merging the responses. > ... ...IIUC in SolrCloud multishard deployments, 9.x's "QUERY./select.requestTimes" is the equivilent of 8.x's "QUERY./select.distrib.requestTimes", while "QUERY./select[shard].requestTimes" is the equivilent of 8.x's "QUERY./select.local.requestTimes" But this has really just attempted to simplify the metric naming -- so that monitoring systems can look at the same info, regardless of wether there are multishard collection/requests or not -- my caveats about how exactly this applies to your original question are still true... > ... > But it doesn't *only* include the "time spent in aggregating the search > results from shards" ... it also includes the time spent determining > which requests to send to which shards, and waiting for the responses to > those (frequently concurrent) requests" -Hoss http://www.lucidworks.com/
