[07:17:47] <pfischer>	 o/
[07:18:09] <dcausse>	 o/
[10:30:52] <dcausse>	 lunch
[10:49:30] <pfischer>	 lunch
[13:18:39] <inflatador>	 <o/
[13:18:47] <dcausse>	 o/
[13:30:45] <ebernhardson>	 \o
[13:31:22] <dcausse>	 o/
[13:34:01] <inflatador>	 .o/
[13:36:52] <inflatador>	 Re: our discussion about building OpenSearch packages, the docs say this is how it's done today: https://wikitech.wikimedia.org/wiki/OpenSearch . Will check w/0lly but it does sounds like an opportunity to use CI
[13:41:36] <dcausse>	 -o znc_libera_dcausse.#wikimedia-search
[14:21:28] <dcausse>	 I think I'm going to regret it at some point but trying to add a notion of "alternative" index to cirrus, shape will be like $basename_$type_alt_$altId_$indexId, e.g. enwiki_titlesuggest_alt_0_1756131652
[14:23:25] <ebernhardson>	 hmm, well one way or another we have to have a second index.  needs some shape :)
[14:55:17] <inflatador>	 interesting, what's the use case for that?
[14:59:06] <ebernhardson>	 inflatador: titlesuggest indices are rebuilt daily (mostly, kinda). to run a/b tests of autocomplete we would need to build a second index that does things differently
[14:59:36] <inflatador>	 oh yeah, that makes sense
[16:20:00] <inflatador>	 quick break, back in ~20
[16:45:59] <inflatador>	 back
[16:51:18] <dcausse>	 dinner
[17:14:58] <ebernhardson>	 hmm, would this work?  Could we inject the start/end anchors at only the ends of source_text, then rewrite ^ and $ as "[" + START_ANCHOR_MARKER + "\n\r\v]"?  https://phabricator.wikimedia.org/T317599#11115894
[17:15:15] <ebernhardson>	 it seems plausible, but something in the back of my head thinks it wont work
[17:15:48] <ebernhardson>	 but maybe i'm just being stubborn for not thinking of it earlier :P
[17:26:06] <ebernhardson>	 being bold and merging the addition of isort+black to discolytics (stacked mr's might be better, but i wanted to get my dumps patch moving and it depends on these): https://gitlab.wikimedia.org/repos/search-platform/discolytics/-/merge_requests/57
[17:30:34] <inflatador>	 lunch, back in ~30 or so
[18:19:35] <inflatador>	 back
[18:23:38] <inflatador>	 headed to the doctor, back in ~90m
[20:19:34] <inflatador>	 back
[20:45:33] <ebernhardson>	 meh..releasing cirrus-reindex-orchestrator fails with git permissions :S will have to poke it a bit
[20:46:10] <ebernhardson>	 oh, weird. The access tokens there are all gone :S happened in another repo before too
[20:58:18] <pfischer>	 ebernhardson: tokens for the CI pipeline to perform git push?
[20:58:51] <ebernhardson>	 pfischer: yea, https://gitlab.wikimedia.org/repos/search-platform/cirrus-reindex-orchestrator/-/settings/access_tokens was empty, didn't even show it as an inactive access token
[20:59:24] <ebernhardson>	 but the actual token value was still stored in CI/CD Variables.  I created a new one and it seems to work now (took two tries, Developer role didn't have push access to protected tag)
[21:00:53] <ebernhardson>	 released now, and kicked off the reindexes i mentioned this morning for the opening_text did-you-mean variant
[21:12:28] <pfischer>	 ebernhardson: Hm, at some point releng had to/wanted to get rid of all tokens w/o an expiration date, but it’s indeed weird that there’s no trace of it
[21:43:26] <inflatador>	 Looks like we'll need to set `vm.max_map_count=262144` sysctl on any kubernetes worker that runs OpenSearch ( https://docs.opensearch.org/2.19/install-and-configure/install-opensearch/index/#important-settings ) . We already set this on our OS hosts, but haven't found the Puppet code yet ;)