[14:02:39] <inflatador>	 <o/
[14:31:29] <ebernhardson>	 \o
[15:03:10] <inflatador>	 <o/
[15:56:10] <ebernhardson>	 streaming updater is stuck, minimal output. I think it's due to some maintenance being done in prod so waiting
[15:56:51] <ebernhardson>	 (i think stuck means messages not coming in)
[15:58:18] <inflatador>	 ebernhardson check mediawiki_security if you're in there
[15:59:18] <ebernhardson>	 ya, i saw
[16:02:53] <inflatador>	 👍
[17:00:41] <inflatador>	 lunch, back in ~40
[17:27:19] <ebernhardson>	 inflatador: when you get back, can we apply the 64kb readahead to opensearch-semantic-search again? Hopefully last time around, just trying to collect some data on even lower memory:index ratios
[17:29:25] <ebernhardson>	 streaming updater came back with the removal of read-only, no action on our part
[17:48:36] * ebernhardson is looking into the deepcat ticket about unexpected results...kinda wish we had a nice way of exposing the raw categories query data.
[18:16:09] <ebernhardson>	 i'm almost certain the answer is it's about category depth limits...but not sure how to present that to the user
[18:21:48] <inflatador>	 back
[18:24:56] <inflatador>	 Github showed me a "semantic search (beta)" thingie today
[18:27:13] <inflatador>	 ebernhardson readahead applied 
[18:27:18] <ebernhardson>	 inflatador: thanks!
[18:28:27] <inflatador>	 np, let me know how it goes
[18:54:18] <ebernhardson>	 i think the deepcat problem was not actually a problem, more a misunderstanding of the user in which categories they were excluding.  No clue how terrible it will be, but curious what comes out. I've asked claude to write a self-contained .html file that would fetch two deepcat &cirrusQueryDumps and show how they differ
[18:54:48] <ebernhardson>	 no way i would have time to put that together myself, but maybe it comes up with something reasonable
[18:57:55] <pfischer>	 Sounds like a good use case. I am curious too. How much context (in terms of folders/sources/docs) did you give claude?
[19:00:21] <ebernhardson>	 i gave it one the json output of one &cirrusDumpQuery.  It did come up with something that shows the diff, the include/exclude for each query, and the intersection, but unfortunately it doesn't really help clarify the problem the user had
[19:13:05] <ebernhardson>	 :q
[20:03:37] <ebernhardson>	 maybe a bit hyper-specialized to this particular report, but interesting:
[20:03:38] <ebernhardson>	 https://people.wikimedia.org/~ebernhardson/deepcat-compare.html?a=-deepcategory%3A%22Maps+of+the+world+by+language%22+deepcategory%3A%222020s+maps+of+the+world%22+-deepcategory%3A%22Wikimania+Map+of+the+world%22&b=-deepcategory%3A%22Maps+of+the+world+by+language%22+deepcategory%3A%222020s+maps+of+the+world%22+-deepcategory%3A%22Wikimania+Map+of+the+world%22+-deepcategory%3A%22SVG+maps+of+
[20:03:40] <ebernhardson>	 the+world+by+language%22+-deepcategory%3A%22English-language+SVG+maps%22+-deepcategory%3A%22Our+World+in+Data+food+and+agriculture+maps+of+the+world%22
[20:03:46] <ebernhardson>	 link too long :(
[20:07:40] <ebernhardson>	 example queries here: https://phabricator.wikimedia.org/T415299#11679002
[20:07:45] <JJMC89>	 !issync
[20:07:45] <ircservserv-wm>	 Syncing #wikimedia-search (requested by JJMC89)
[20:07:46] <ircservserv-wm>	 No updates for #wikimedia-search
[20:09:27] * ebernhardson is actually mildly impressed with that result
[20:44:03] <ebernhardson>	 hmm, opensearch-semantic-search is getting EOF's from inference.discovery.wmnet
[20:45:35] <ebernhardson>	 Received error from remote service with status code 502, response headers: {content-length=[53], content-type=[text/plain; charset=utf-8], date=[Thu, 05 Mar 2026 20:37:12 GMT], server=[istio-envoy], x-content-type-options=[nosniff], x-envoy-upstream-service-time=[1]}
[20:45:48] <ebernhardson>	 opensearch-semantic-search-masters-0] Remote service returned error: Error from remote service: dial tcp 127.0.0.1:8080: connect: connection refused with status: BAD_GATEWAY
[21:40:28] <ebernhardson>	 pfischer: semantic search should be deployed, but due to ^^ i can't verify if it's working. The train did roll forward so in theory if kserve starts responding it should work. Probably. Filing a ticket for the kserve proble
[22:07:07] <ebernhardson>	 filed https://phabricator.wikimedia.org/T419174
[22:55:34] <ebernhardson>	 sigh..checking things out there is a part that didn't occur to me, interwiki search on frwiki means semantic wont work (yet) on Special:Search, has an issue loading the profile cross-wiki.  API will still be fine though
[22:58:20] <ebernhardson>	 sigh, actually it wont. I forgot that the opensearch clusters requires auth to allow ml predictions ...
[23:01:57] * ebernhardson puzzles around in the opensearch security settings...
[23:08:55] <ebernhardson>	 ok, created an ml_predict_role in _plugins/_security/api/roles/ml_predict_role,  and then a rolemapping at _plugins/_security/api/rolesmapping/ml_predict_role with the end result of assigning cluster:admin/opensearch/ml/predict to opendistro_security_anonymous_backendrole
[23:11:47] <ebernhardson>	 running out of time today, but i'll figure out how to get that into the cirrus-toolbox opensearch_config.py tomorrow
[23:16:47] <ebernhardson>	 alternatively if we care we should be able to create a proper user, but i guess just out of momentum from how things already work...i'm fine with not having cirrus auth with the cluster.