[15:54:46] <ori>	 why does Wikimedia\Rdbms\DatabaseMysqli::mysqlConnect take up ~4.5% of index.php wall-clock time? that seems like a lot
[16:05:39] <Krinkle>	 ori: half of it is from post-send deferreds which I"m guessing either re-opens ones we closed or opens additional ones.
[16:08:22] <Krinkle>	 the other ones are: one for sqlbag (parsercache), one for probably the local wiki (typcally first triggered by WikiPage in early MediaWiki.php code), one for wikidata triggered first by Wikibase, and sometimes one more to the local wiki primary DB which seems to be triggered by a cache miss on isReadOnly.
[16:09:43] <Krinkle>	 from a random XHGui run on commonswiki/index.php?action=raw which does only 2 connnects: https://performance.wikimedia.org/xhgui/run/symbol?id=634d562ed148c5632f0598fb&symbol=Wikimedia%5CRdbms%5CDatabaseMysqli%3A%3AmysqlConnect - reports as 2ms inclusive walltime.
[16:11:47] <Krinkle>	 I dont' know offhand if ~1ms per connect is high or not historically
[16:12:13] <Krinkle>	 appserver latencies tend to be 100-400ms at p75 overall.
[16:12:23] <Krinkle>	 some endpoints <50ms 
[16:12:34] <Krinkle>	 like opensearch, load.php and some rest.php calls
[16:14:46] <ori>	 does excimer differentiate between request-blocking and post-send deferreds?
[16:37:55] <Krinkle>	 ori: yes, in so far that they have different cal stacks. restInPeace/postOutput is post send.
[20:00:42] <ori>	 neat. thanks.