[15:10:26] <Krinkle>	 TimStarling: https://logstash.wikimedia.org/app/kibana#/doc/logstash-*/logstash-deploy-2019.12.01/mediawiki/?id=AW7BBYdjh3Uj6x1zeKm6
[15:10:40] <Krinkle>	 this one is also action=opensearch, but that's not fast/cheap per se
[15:10:48] <Krinkle>	 on this one, it failed whilst executing Lua
[15:11:23] <Krinkle>	 It's because of TextExtracts which has a cold case in which it creates a ParserOutput object
[15:13:26] <Krinkle>	 Scrap that, it doesn't even use the parser cache. I don't know why not.
[15:16:49] <Krinkle>	 Every single minute,  5% of opeonsearch backend hits takes 10s or more to complete. with 1% at a minute or more (presumably times out)
[15:16:49] <Krinkle>	 https://grafana.wikimedia.org/d/000000559/api-requests-breakdown?orgId=1&refresh=5m&var-metric=p95&var-module=opensearch
[15:21:53] <Krinkle>	 and for cache misses (trigger by entering a somewhat unusual query), responses take 20-30s for me indeed (commonswiki). With XHGui showing > 90% spent in TextExtracts\ApiQueryExtracts::parse
[16:59:00] <Krinkle>	 Filed https://phabricator.wikimedia.org/T240691
[16:59:10] <Krinkle>	 recommending undeployment in the short-term.