[08:22:14] <jayme>	 ottomata: https://wikitech.wikimedia.org/wiki/Kubernetes/Resource_requests_and_limits - we have to run more tests with envoy as of now. Is it a "real" problem (like in latency increase) or is it "just about the metrics"?
[09:54:57] <elukey>	 it is being throttled heavy (tens of ms) so latency is likely affected, but so far I don't think we see holes in graphs etc..
[09:55:08] <elukey>	 https://phabricator.wikimedia.org/T347477#9299688
[14:28:02] <ottomata>	 jayme: good q, not sure, but mostly i'm wondering why throttling at all, if not anywhere close to limits?
[14:28:45] <jayme>	 please check on wikitech if that maybe answers that question
[15:27:22] <ottomata>	 okay, i sorta understand that if there a buncha threads scheduling can get divvyed poorly and cause throttling
[15:27:38] <ottomata>	 i think we may be affected, trying something...
[15:40:18] <jayme>	 yay, success :D If you want more, you may want to read all/some of the external resources. But it's not exactly fun
[15:52:02] <ottomata>	 i have skimmed some, understanding deeply would take ummmmm more time than i have right now :D
[17:24:30] <ottomata>	 fyi, we set service-runner num_workers to 0, meaning there is only one nodejs process spawned, running master+worker in same process.  This means 1/2 the number of threads created (UV_THREADPOOL_SIZE is set to 128 in the docker image).  
[17:24:35] <ottomata>	 since then: no throttling!
[17:24:47] <ottomata>	 https://grafana.wikimedia.org/goto/6KmODK4Sk?orgId=1
[17:25:06] <ottomata>	 https://phabricator.wikimedia.org/T347477#9302664
[17:45:50] <inflatador>	 {◕ ◡ ◕}