[10:31:11] <tgr>	 Krinkle: I was trying to correlate yesterday's ResourceLoader 5xx alert with Logstash: https://grafana.wikimedia.org/d/000000066/resourceloader?orgId=1&from=1712862062090&to=1712881502090&viewPanel=56
[10:31:22] <tgr>	 but mediawiki-errors has nothing: https://logstash.wikimedia.org/app/dashboards#/view/mediawiki-errors?_g=h@46e3fbe&_a=h@8141eda
[10:31:51] <tgr>	 am I wrong in assuming that a varnish 503 means the backend timed out and there should be a corresponding error in the MediaWiki logs somewhere?
[10:32:36] <tgr>	 (there was a small DDoS attack so it's no mystery what caused it, I'm just being curious)
[10:58:59] <mszabo>	 AIUI a timeout due to queueing will not necessarily result in an error on the PHP side
[11:12:43] <tgr>	 as in, fcgi or apache or something like that discards the request after a certain time because there's no available worker, and PHP does not get involved at all?
[14:06:36] <Krinkle>	 tgr: I believe 503 is exclusively for when the request never reaches the appserver. Not exactly lack of worker at Apache level, but rather ATS proactively refusing to try (there is a configured limit to how many connections varnish/ATS will send to each backend).
[14:09:30] <Krinkle>	 In practice, the only 5xx from php are 500. Including for timeouts at that level.
[14:11:14] <Krinkle>	 In theory timeouts can at higher layers but that suggests a failure to enforce timeouts at lower layers. https://wikitech.wikimedia.org/wiki/HTTP_timeouts#App_server