[08:53:13] ebernhardson: very nice, thanks for putting this together! [11:05:55] lunch [14:14:14] o/ [14:45:44] \o [14:46:28] o/ [15:22:21] ebernhardson: I will definitely take a look and make minor edits or ask questions if need be. [15:23:00] Do we want to push the Wednesday Meeting back an hour for the P&T meeting, or is everyone okay with the recording? [15:23:23] hmm, given recent things it seems like maybe P&T will be something we should do? [15:23:48] also, thanks! [15:29:45] Hm, I invited Research for todays Wednesday Meeting to share their progress on the Golden Set. But I can understand if you would want to join P&T. [15:34:48] nah, if we have visitors i can do the video. [15:35:05] ebernhardson: thanks. dcausse: how about you? [15:35:35] pfischer: happy to join the wed meeting [15:54:30] yeah, recording is fine if we have visitors [19:42:57] ebernhardson: dcausse: Could you please have a look at T414623? Please feel free to add details [19:42:58] T414623: [Vector Search] Estimate resource consumption at scale - https://phabricator.wikimedia.org/T414623 [19:43:52] sure [19:44:12] i'm just trying to write out now what limitations we have for setting up relforge<-->mediawiki, i wonder how much tech debt we want to take on [19:44:21] Things like no LVS, no TLS (which means no shared name) [19:44:37] but those things, historically, are time consuming [19:45:24] Hm, is there any alternative? [19:45:49] the alternative is plain http, and putting a list of hosts in the CirrusSearch config. It won't gracefully handle a server-down [19:46:44] ebernhardson: since no round-robin would happen since no reverse proxy takes care of it? [19:47:01] right, we would just be taking a random host from the server list and hoping it works [19:47:12] it's all possible, but it means redirecting some SRE effort, we can't do those bits [19:49:03] I guess that’s fine, the client should handle response errors like this. [19:53:32] A couple of years ago resilience was a thing and I remember libs for Java and JS popping up, I would be surprised if there isn’t one for PHP that handles retries, for example. [19:55:27] ooh, thats right. Usually we have envoy, i guess i better note that as well [19:55:46] not sure how envoy plays with the host-list thing [19:55:57] (or add more tech debt by skipping envoy too) [19:56:34] envoy might have something, but it's going to be mixed into the helm templates used by everyone so there might be pushback on doing it "wrong [20:03:03] hm, afaik we’d need sth. like a service for an envoy port mapping. I’ll reach out to SRE. Maybe its possible to have a list of hard coded host names there, too. [20:03:48] Alright, I am out for today! See you around. [20:16:24] ebernhardson: don't have full context but happy to have a quick chat about envoy configuration if it'd help [20:17:30] cdanis: sure! the general idea is we have a cluster that was always used for internal testing, living in the analytics network, but now the ask is to run a live-test with mobile app users that has to query that service. So basically turning a test service into a production service (or at least, a facsimile of one) [20:17:50] I'll ping you on the ticket (T413969) right now i'm mostly collecting requirements [20:17:51] T413969: Make semantic search accessible through Action API - https://phabricator.wikimedia.org/T413969 [20:18:21] hmm, maybe fascimile is the wrong word, but it sounds nice :P [20:18:44] makes sense, thanks! [21:34:42] ryankemper https://gerrit.wikimedia.org/r/c/operations/alerts/+/1226282 CR opensearch on k8s alerts if you have time to look [21:39:37] cdanis ebernhardson can y'all include me in these discussions too? I'm curious if there's a something we could do with envoy/istio without having to stand up an LVS pool, ref T404119 [21:39:37] T404119: Explore options for service discovery at WMF - https://phabricator.wikimedia.org/T404119 [21:51:12] hmm, the awkard thing about putting a plan in a phabricator comment is no-one else can edit it. But it doesn't always feel right in the ticket description, and putting it in a paste renders terribly... [22:03:05] inflatador: 5’ for pairing [22:03:48] ryankemper ACK, I'm around