[07:01:28] the only nested queries running on the main cluster are from GeoData, it caused issues in the past... [07:17:15] sigh fulltext qps is now 9k/s at the lowest and past 1k/s some times... [07:35:57] the automated poolcounter never rejects anything, wondering if something changed at the traffic layer such that we no longer properly flag those... [10:13:30] semantic-highlighting-staging deployed to liftwing, running some benchmarks and will try to get an opensearch search pipeline using it to have a full end-to-end test [10:13:41] lunch [13:15:59] o/ [13:41:16] \o [13:41:36] 9k/s!! [13:42:09] oh, bottoms above 900 [13:42:30] oops sorry yes, 900 - 1k/s [13:42:51] latency is down though, suggesting those are cheap-ish queries [13:43:28] but yea, suggests some analysis of what the volume is might be needed to shift it to the automated bucket [13:43:54] yes, started looking into those, haven't anything except suspicious traffic to zhwikisource with the same pattern we've, hitting the search endpoint to fetch images and thus hitting the commons index [13:44:13] sigh can't type [13:47:09] the plugins deb is also published to apt.wikimedia.org now, should be able to publish the 3.8.0 image as well today [13:48:27] nice! [14:10:37] dcausse sorry I lost track of time. I don't really have anything for pairing but LMK if you need anything [14:11:04] inflatador: np! don't have much either [14:38:51] I'm gonna roll-restart CODFW semantic search to test Balthazar's script, assuming it works I should have an MR for cirrus_toolbox soon [14:45:03] semantic-highlighting-staging seems to be working https://en.wikipedia.org/w/index.php?search=how+old+is+the+universe%3F&title=Special%3ASearch&profile=advanced&fulltext=1&ns0=1&limit=3&cirrusSemanticSearch=hl [14:45:37] also seems to have the nice same multi-lingual properties: https://fr.wikipedia.org/w/index.php?title=Sp%C3%A9cial:Recherche&cirrusSemanticSearch=hl&fulltext=1&limit=3&ns0=1&profile=advanced&search=how+old+is+the+universe%3F [14:54:50] 👀 [14:56:38] That’s awesome. Great work! 🎉 [14:59:12] dcausse: That’s worth a follow-up post on https://wikimedia.slack.com/archives/C0975D4NLQY/p1789456401247699?thread_ts=1785263959.247429&cid=C0975D4NLQY [15:00:02] pfischer: sure, will deploy the non-staging thing first [15:25:57] oooooh. nice [17:07:57] as promised, MR up to add the pod restart script to toolbox https://gitlab.wikimedia.org/repos/search-platform/cirrus-toolbox/-/merge_requests/15 [18:44:44] meh, cirrussearch-opensearch-image is pulling mirrors.wikimedia.org on apt-get update...must need a change to the base image [20:55:03] ebernhardson I'm working on T438238 , just wondering what is a plausible way to look for fulltext queries in turnilo. I'm using the URI query regex `^\?action=query?` but LMK if that doesn't work [20:55:04] T438238: Cirrussearch: investigate unexpected rise in fulltext web searches - https://phabricator.wikimedia.org/T438238 [20:55:46] Dunno if you can long to turnilo but https://w.wiki/Udv6 is what I've got so far [20:55:47] inflatador: hard to say. For api.php you can probably look for &srsearch= or &gsrsearch=. For mw-web it's harder, the urls are all translated [20:56:04] i haven't looked in turnilo, maybe they have the x-analytics-map which iirc has the un-translated name for special pages [20:56:51] looking for `[?&]search=` is probably close enough i guess [21:11:11] I can't go back too far in turnilo without making druid squeal, but I don't see anything obviously bot-generated so far. Not that that means a lot