sarkaramrit1993 commented on PR #5238: URL: https://github.com/apache/solr/pull/5238#issuecomment-6541381635
Batching is in [92a4aa0](https://github.com/apache/solr/pull/5238/commits/92a4aa008f12d64503bfd1c7d53e0cae0b5719dc). The listener thread now drains the queue in batches (`take` + `drainTo`, then a 1 ms pause) instead of waking once per search. JMH SampleTime, single node, 4 threads, `q=id:0`, 3 forks, about 2.5M searches per fork (µs): | | no listener mean | no listener p50 | no-op listener mean | no-op listener p50 | |---|---|---|---|---| | before batching | 75.5 | 70.7 | 79.1 | 75.8 | | after batching | 73.4 | 70.1 | 74.7 | 70.5 | Before batching, a no-op listener added about 5 µs at p50. After batching, the gap is smaller than the fork-to-fork noise on this laptop (a few µs between forks of the same config), so there's no measurable overhead. These are closed-loop runs, so I'm not reading anything into p99. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
