[09:55:16] <inflatador>	 .o/
[09:56:16] <pfischer>	 o/ Are you at the offsite? It’s super early for TX?!
[11:00:10] <inflatador>	 I'm attending virtually. I was all set to go and had a conflict at the last minute
[11:08:50] <dcausse>	 power's finally back...
[11:08:52] <dcausse>	 o/
[11:09:40] <dcausse>	 ebernhardson: sounds great, thanks! :)
[14:31:53] <ebernhardson>	 \o
[14:33:04] <ebernhardson>	 dcausse: i suppose the only warning is it is a vegetable-forward place. Turkish flavors, but not as meat heavy as a more traditional turkish place.  Some reviews mention "chlorophyll" flavors. I suppose call it a bit experimental
[14:33:22] <dcausse>	 o/
[14:54:27] <inflatador>	 <o/
[14:59:17] <dcausse>	 ebernhardson: totally fine by me :)
[15:06:43] <ebernhardson>	 ok cool, reservation set wed 19:00
[15:09:44] <dcausse>	 thanks!
[15:12:18] <pfischer>	 Trey314159: ebernhardson: How do you feel about skipping the Retro today? (dcaussse is fine either way)
[15:15:13] <ebernhardson>	 fine by me
[15:18:58] <Trey314159>	 that's fine
[16:06:17] <pfischer>	 Semantic Search folks raise a question: Would Cloud VPS be an alternative to waiting for the k8s OpenSearch cluster? It is my understanding, that basically, that leaves us in the same spot as with relforge: We’d have to orchestrate the OS cluster + LB + TLS termination. So no gain. What do you think ebernhardson and dcausse?
[16:07:12] <dcausse>	 seems unlikely to me that we could access cloud vps from mw?
[16:07:57] <pfischer>	 yep, Antoine mentioned that, separate private network
[16:09:12] <pfischer>	 How about egress to public internet from MW? TIL: we could also get a hosted OS cluster (AWS?). There’s budget for that.
[16:10:53] <inflatador>	 pfischer we have a working OpenSearch 3 image and I already have a cluster deployed with it in K8s
[16:11:48] <dcausse>	 not sure we want user queries to leave the cluster and I doubt we can make mw dependent on external resources, it'd have to be the android hitting that external resource not mw
[16:13:58] <ebernhardson>	 yea from a prod perspective, we aren't supposed to ever call cloud from prod.  It does happen, but it's always been considered inappropriate.  For the external cluster i suspect similar, maybe it would be ok but i'm not sure how our privacy policy looks like there.
[16:16:06] <ebernhardson>	 i also worry about the idea that " needing to know if  we need to pivot to something else other than Semantic Search as a feature".  If they roll it out and it's mid...is that a fundamental problem, or an iteration problem. Personally i would not expect great results on the first try.
[16:17:00] <ebernhardson>	 maybe good enough, will see
[16:18:01] <inflatador>	 As far as the OpenSearch k8s cluster, we're waiting on https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/1230512 . If everyone's cool with the namespace, we can merge this and probably deploy next week
[16:18:05] <inflatador>	 cc brouberol 
[16:18:32] <ebernhardson>	 ooh naming, the thing we think about for 12 minutes, and then live with the consequences for 12 years :)
[16:18:37] <ebernhardson>	 looking
[16:19:07] <inflatador>	 ebernhardson I think you said it was OK, but I wanted to let others in the Slack thread respond. It's been 4 days so I'm fine with pushing ahead whenever
[16:19:14] <ebernhardson>	 yea i'm fine with opensearch-semantic-search
[16:20:38] <ebernhardson>	 inflatador: maybe not worth looking into, but i wonder about the min/max memory.  naively i would set them exactly the same, but i'm not entirely sure how the linux disk-cache interacts there.
[16:21:10] <inflatador>	 OK, once the offsite activities are done I'll merge all this stuff and try to deploy. Looking at T413969 it seems like there's also a need for building OpenSearch 3 plugins too
[16:21:11] <stashbot>	 T413969: Make semantic search accessible through Action API - https://phabricator.wikimedia.org/T413969
[16:21:57] <inflatador>	 yeah, we'll def set min/max the same. If you set them too far apart it leads to premature thrashing, we've seen that with cgroups on the stat hosts
[16:41:27] <pfischer>	 inflatador: Thank you! I am fine with that name too. Good to know that we are almost there already. All I had so far was a commitment to end of February.
[16:57:45] <brouberol>	 inflatador: I +1ed all patches this morning, so feel free to deploy everything
[16:58:27] <brouberol>	 pfischer: i.nflatador figured out that, contrary to what is advertised, the operator 2.7 (that we currently run) manages to deploy opensearch 3
[17:02:57] <ebernhardson>	 nice! 
[17:06:43] * ebernhardson is being terribly indecisive on how to flip FullTextResultsType to SemanticResultsType
[17:06:57] <ebernhardson>	 we currently just always set it in Searcher::search
[17:07:29] <ebernhardson>	 maybe just let semantic builder override it, but seems awkward
[17:20:27] <ebernhardson>	 awfully suspicious filepath from `mw docker custom where`: /home/ebernhardson/.config/mwcli/mwdd/default/home/ebernhardson/.config/mwcli/mwdd/default/custom.yml
[17:22:16] <inflatador>	 for the k8s max/min memory stuff, those are actually namespace defaults as opposed to the requests/limits that will be applied to opensearch pods
[17:24:01] <inflatador>	 we have to do that b/c the bootstrap pod is a full-fledged opensearch node that needs at least 2 GB to stand up the cluster
[17:24:33] <inflatador>	 and there are no other levers to pull to effect bootstrap pod resources
[17:25:02] <inflatador>	 **affect**