Amrit Sarkar created SOLR-18629:
-----------------------------------
Summary: cpuTime leaves out shard CPU when a query limit ends a
distributed search early
Key: SOLR-18629
URL: https://issues.apache.org/jira/browse/SOLR-18629
Project: Solr
Issue Type: Bug
Reporter: Amrit Sarkar
With {{solr.log.cputime=true}}, a distributed search reports the shards' CPU
time in the {{cpuTime}} response header. When a query limit ({{timeAllowed}},
{{cpuAllowed}}, {{memAllowed}}) stops the coordinator after some shards have
answered, {{SearchHandler.processComponentsDistrib}} returns early and never
adds it. {{RequestHandlerBase}} then fills {{cpuTime}} with the coordinator's
own CPU time, so partial results look much cheaper than they were.
Fix: publish the shard CPU collected so far on each early return, as long as at
least one shard request finished.
A shard that throws without {{shards.tolerant}} has the same gap. Not covered
here.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
