[08:30:57] good morning! [08:33:55] atsukoito: hi! [08:34:23] atsukoito: the ttmserver-export fix is being shipped, should be avalaibe in a couple minutes [08:36:12] dcausse: thanks! i'll look for some small wiki to test it on before committing to full re-index [08:36:37] I'll drop the index and re-index wikifunctions [08:36:56] and then will test from shell [08:37:33] sounds good [08:44:56] atsukoito: deployment done [08:50:36] i'll start re-indexing [09:16:24] dcausse: i reindexed wikifunctionswiki and `query("de", "en", "Vorheriges Update")`now returns the results [09:16:33] i'll proceed with re-indexing [09:16:51] atsukoito: awesome! [09:58:55] mw dev images now fails because of the removal of mirrors.wikimedia.org, trying bookworm-php84-fpm:1.0.0-s2... [09:59:51] T423972 [09:59:51] T423972: Rebuild dev-images using a base image without mirrors.wikimedia.org in the apt sources - https://phabricator.wikimedia.org/T423972 [10:01:04] lunch [12:50:01] \o [12:51:13] o/ [12:51:38] ebernhardson: hii! [12:54:25] cindy is going to fail all patches til we newer images (T423972) [12:54:26] T423972: Rebuild dev-images using a base image without mirrors.wikimedia.org in the apt sources - https://phabricator.wikimedia.org/T423972 [12:57:13] fun :S [13:02:51] o/ [13:09:18] what should we call the cirrus html helper tools repo/subdomain? I have it in a local dir cirrus-html-helpers but that name seems meh :P cirrus-tools kinda overlaps with cirrus-toolbox [13:11:43] ebernhardson: yes cirrus-toolbox is mainly for backend utilies so probably its own project makes sense, for the name that's always the hard part, unsure if "html" is necessary... if it's all about "explaining" perhaps "cirrus-explain" or the like would make sense? [13:12:34] if it's more than just explaining I'm not so sure [13:13:05] it has a few tools, right now it's the morelike, deepcat-compare, templated insource-search, and cirrus-explain tools [13:13:51] what's the templated insource-search? [13:14:51] dcausse: it's an attempt at a "documented" way of running regular expressions. Has expressions with templated variables, and then fields for collecting those variables. [13:15:22] basically it takes a .json file (that could be per-wiki hosted) to template out a bunch of form fields. not sure if it's super useful [13:15:47] if it's more than explaining... a broader name could be "cirrus-companion-tools" but well I'm bad at naming things... [13:15:53] mostly this: https://people.wikimedia.org/~ebernhardson/insource-search.html [13:16:31] hmm, companion isn't terrible [13:17:37] the regex templates is a nice, I kind of forgot you built that [13:17:47] s/nice/nice idea/ [13:18:02] yea i think there is something to that general idea, although not sure this is exactly it [13:30:42] ouch I have a spark job stuck for 13days :/ (https://yarn.wikimedia.org/proxy/application_1778834239566_429889/) [13:31:12] meh... the ui reports it's done tho... [13:33:00] ah it's because it's looping waiting for the index to go green... haven't added any timeouts there :/ [13:35:01] disk usage is close to 90% on all nodes :/ [13:41:32] dropping a stale index and will perhaps decrease replicas for enwiki on opensearch-semantic-search.svc.codfw.wmnet to help with the new index being built [13:48:19] :S [13:48:25] i suppose fine though [14:23:38] back to green [14:26:51] only 53 queries over 7days are using -prefix:, some are due to dumb filtering on my spark jobs but some seems legit [14:27:27] few have a namespace like -prefix:user:foo [14:27:42] wondering if the query logs have the original set of namespaces [14:27:59] i thought i couldn't do worst names, but then claude came up with `cirruscope` [14:28:23] :) [14:28:35] Nik would probably say it's "cute" :) [14:28:39] what about cirrus-lens.wmcloud.org? [14:28:51] +1 [14:29:10] will go with that, seems plain enough [14:53:39] sigh... yes some -prefix seems to rely on the fact that it switches the ns: got Special:Search URLs with default params but the query looks like "filetype:audio -prefix:file:fr" [14:55:02] well if it's on commons ns_file should be searched on by default so hopefully less of a breaking change [15:15:24] https://cirrus-lens.toolforge.org/ [15:15:37] nice! [15:51:04] Trey314159 what is the Slack channel you mentioned? [15:51:25] other randomly interesting thing, i asked claude to spawn a tree of sub-agents to review cirrus for settings that aren't used anymore, it came up with a couple: https://etherpad.wikimedia.org/p/cirrus-settings-audit [15:51:46] inflatador: #linguists-guild [15:51:53] and then asked it to use mediawiki-config history to find config that is basically static and could potentially turn from a setting into a constant: https://etherpad.wikimedia.org/p/cirrus-settings-simplification [15:52:09] not entirely sure what to do with them yet, the first doc can probably review and remove but it's only ~6 settings. [15:52:25] but the second found ~25 settings it thinks could turn into constants [15:53:26] was also just nifty though to ask it to use a tree of sub-agents, and it setup a workflow with a controlling agent, then sub-agents which reviewed ~8 settings each, and then a followup dedicated agent that verified each candidate flagged by the first pass [15:53:47] sub-agent per setting verification [15:53:52] ebernhardson: seems like these etherpads are empty [15:54:05] hmm :S i see them locally...sec [15:54:17] maybe i should use phab, it's juts i was logged out and have to grab my phone for the two-factor auth...sec [15:57:07] dead settings: https://phabricator.wikimedia.org/P94327 [15:57:41] potentially constants: https://phabricator.wikimedia.org/P94328 [15:58:51] * ebernhardson wishes phab paste would render markdown [16:02:31] these were surprisingly quick to generate, maybe 30 min (and used north of a million tokens for each) [16:05:52] yes, sad that phab only supports markdown in comments but not in paste [16:06:34] used gitlab gist for this but it has the same annoyance, you have a description that is markdown but then you must upload the gist content [16:06:37] is it worth some time to review and implement? If we could shave 30-ish settings it helps...but there would still be 150 settings [16:06:58] I think it might be worth it but that's a lot to unpack [16:07:23] some settings are definitely not used and were added "just in case", like some of the *Profiles one [16:07:24] indeed we can't just delete things because it says so, requires some per-setting consideration [16:07:46] the typos should definitely be reviews [16:09:09] like CirrusLanguageLanguageKeywordExtraFields should definitely be fixed in extension json [16:09:44] it's what you get when SearchConfig does not scream for unknown settings :/ [16:09:53] core's Config is much stricter [16:10:24] also core settings are all php constants [16:10:43] oh, yea that's maybe something claude can do. I've pondered that but didn't want to go around doing all the mechanical work [16:10:53] true [16:16:00] did not stop cindy yes, seems to actually run with bookworm-php84-fpm:latest, will see [16:16:05] s/yes/yet [16:52:43] Trey314159 ACK thanks [17:10:42] dinner [17:24:48] i guess it did get used: https://github.com/wbstack/mediawiki/blob/main/dist/wbstack/src/Settings/LocalSettings.php#L714