[09:11:29] <pfischer>	 dcausse: I was about to summarize our weekly report, but I am not permitted to create the following page: https://wikitech.wikimedia.org/wiki/Search_Platform/Weekly_Updates/2023-10-18 - are you permitted?
[09:12:10] <dcausse>	 pfischer: this seems strange, looking
[09:13:03] <dcausse>	 pfischer: do you have an error message? or perhaps you're not logged-in?
[09:13:41] <dcausse>	 I think wikitech does not allow anon edits
[09:15:04] <pfischer>	 I am logged in and I don’t see an error
[09:15:57] <pfischer>	 Ah, maybe it has to do with account migrations… looking
[09:45:00] <dcausse>	 errand+lunch
[12:09:51] <gehel>	 Reminder: engagement & inclusion survey is out. The deadline is Nov 1, just after I'm back from vacation. Please take time to answer it!
[13:21:42] <inflatador>	 <o/
[13:23:00] <dcausse>	 o/
[14:00:51] <ebernhardson>	 \o
[14:16:21] <dcausse>	 o/
[14:18:58] <ebernhardson>	 i wonder if something about categories in wdqs broke? In the ticket about deepcat user mentioned now deepcategory:Science is simply returning incategory:Science. That's usually what the result of the rdf query looks like when the db is empty
[14:19:43] <ebernhardson>	 re: T369808
[14:19:44] <stashbot>	 T369808: The Commons search "deepcategory" operator often does not work (Deep category query returned too many categories) - https://phabricator.wikimedia.org/T369808
[14:19:49] <dcausse>	 perhaps? we don't have much monitoring on this endpoint
[14:21:11] <dcausse>	 wdqs2020 is having few triples than other nodes (https://grafana-rw.wikimedia.org/d/000000489/wikidata-query-service?orgId=1&refresh=1m&var-cluster_name=wdqs&var-graph_type=%289103%7C9194%29)
[14:22:12] <dcausse>	 but maybe cirrus targets the internal endpoint
[14:22:57] <dcausse>	 for which all nodes are varying triples count :/
[14:23:20] <ebernhardson>	 looking at the public side tof the api, getting reasonable responses to the documented example queries
[14:23:31] <ebernhardson>	 hmm, i'm not sure which cirrus queries. checking
[14:23:55] <dcausse>	 well +/- 400k triples
[14:24:49] <inflatador>	 I did start and stop categories a couple of times on wdqs2020 a couple of weeks ago when I was looking at the migration stuff, but nothing too disruptive AFAIK
[14:25:10] <ebernhardson>	 heh, we set wgCirrusSearchCategoryEndpoint twice in the same config array :S should fix. but i think the second wins, which means http://localhost:6009/bigdata/namespace/categories/sparql and 6009 is wdqs-internal
[14:25:24] <dcausse>	 :/
[14:26:33] <dcausse>	 readiong the code there should be a warning
[14:27:52] <ebernhardson>	 hmm, well i suppose this isn't super important, was just something i thought of reading the morning emails :)
[14:28:39] <ebernhardson>	 but i'll look into it a little today
[14:30:43] <dcausse>	 I think it's because media search does not display the warning
[14:30:50] <dcausse>	 https://commons.wikimedia.org/w/index.php?go=Go&search=deepcategory%3AScience&title=Special:Search vs https://commons.wikimedia.org/w/index.php?go=Go&search=deepcategory%3AScience&title=Special:MediaSearch&type=image
[14:31:13] <ebernhardson>	 oh, that makes sense. We should get that added
[14:32:20] <dcausse>	 also the timeout at 3sec is perhaps a bit restrictive
[14:33:07] <ebernhardson>	 i just adjusted the example query from docs to query Science, times out on the public api too :P But yea maybe 3s could be higher
[14:33:40] <dcausse>	 ouch even with a limit ?
[14:34:37] <ebernhardson>	 oh, i guess the example subcategories query is different than the one in the keyword, it has maxIterations 8 instead of a depth limit
[14:34:47] <ebernhardson>	 but i'm assuming that results in the same thing
[14:36:25] <ebernhardson>	 i wonder how we could "know" when to limit the depth. In a repl test i can get results with depth=3 on Science, but how would we know when to limit
[14:37:22] <ebernhardson>	 or maybe with a 3s timeout we could simply fallback from 5 to 3 and hope it's better
[14:37:39] <ebernhardson>	 but that might be mysterious if query behaviour depends on backend load
[14:37:55] <ebernhardson>	 i dunno, anyways i gotta do a school run, back in 20
[15:00:23] <ebernhardson>	 back
[15:02:34] <dcausse>	 switching the depth seems confusion... I realize that in its current form it's also confusing because you may return a number of categories < thrshold and not display any warnings but just because you did not scan the full graph...
[15:02:43] <dcausse>	 s/confusion/confusing/
[15:06:25] <ebernhardson>	 also that the current error cases don't line up.  Too many categories -> filter is empty. Way too many categories (times out) -> filter is the source category
[15:07:11] <ebernhardson>	 maybe we should let users provide a depth and the "smart client" can choose a better depth
[15:07:21] <ebernhardson>	 basically punt :P
[15:13:02] <dcausse>	 true, yes peyhaps leaving the choice of maxdepth to the user could be interesting indeed
[17:10:35] <dcausse>	 heading out, have a nice week-end