[08:23:38] fulltext qps dropped down to 650 during the night... [08:54:54] heading out (most likely out for the whole day) [13:33:04] Seddon created a SCROLL request for Semantic Search: https://wikitech.wikimedia.org/wiki/SCROLL/Semantic_Search_Phase_2 [13:33:53] A lot of it is going to need input from Search Platform team. This isn't urgent yet. dcausse, ebernhardson: could you have a pass on it, see what we already have and what we need to figure out? [14:06:52] \o [14:30:54] gehel: sure i'll take a look [15:50:02] a fun term for claude-isms: the Claudish robolect [18:44:43] * ebernhardson sighs...tried to cut a 2.19.5-wmf3 release and...i forgot the javadocs patch :P one more try... [19:55:47] looks like the CPU pressure finally bottomed out in CODFW: https://grafana-rw.wikimedia.org/goto/svkxcn?orgId=default . Checking percentiles to see if P95 and friends also went down [20:36:20] oh fun, it looks like the new itermittent failures in cindy are due to newer chrome. Locally i get 20 out of 20 passes with chrome 140, but then updating to chrome 156 fails 5 of 20 [20:39:52] for the cpu pressure, looks like eqiad came back online [20:40:32] probably about 15:20 [20:58:19] will deploy the revert so we stop forcing more_like to eqiad [21:01:22] Yup. I'm trying to organize my thoughts on that CODFW vs EQIAD ticket, a bit sidetracked with the requestctl stuff [21:03:06] If we wanted to target search URIs with a requestctl rule, is `^/(?:w/index\.php\?(?:.*&)?search=|wiki/Special(?::|%3A)Search(?:/[^?]+|\?(?:.*&)?search=))` a reasonable regex? [21:03:26] that's what Claude was "thinking" [21:14:47] +1 on the morelike patch [21:16:18] it's tough, the urls are translated into the local language :( [21:16:58] maybe something like `[?&](g?sr)?search=` ?hmm [21:17:13] capturing either the search, srsearch, or gsrsearch parameters [21:50:13] remind me, are those bot UAs generating the most entries in cirrus request logs or are they doing expensive queries? [22:47:12] nm, found that