[10:01:37] lunch [13:07:09] greetings [13:16:30] yay, it looks like notifications are working today [14:40:09] Appt’s taking a bit longer than expected, will be 30 mins late to weds meeting [15:12:09] @team: Wednesday meeting is open: https://meet.google.com/eki-rafx-cxi (cc: ejoseph, others?) [15:15:11] o/ - Archive indices for cirrus: it looks to me like these are made (but unreadable) even when `CirrusSearchEnableArchive` is disabled is that expected? Or is this setting for a different thing? [15:24:20] inflatador: might be interesting for the tmux script your were working on for monitoring elastic: https://github.com/tmuxinator/tmuxinator [15:24:29] there is even a .deb package available [15:28:11] gehel ah yes, I've been meaning to check out tmuxinator for awhile [15:28:30] honestly, I don't have experience with it. But been meaning to check it as well [15:57:31] tarrow: hmm, it worked years ago when we put it together in a hackathon but i haven't looked recently, it should tie into mediawiki core through a hook that ends up replacing an sql query [15:57:33] errand, back in ~30 [15:59:32] tarrow: i guess it's not a hook, it's SearchEngine::searchArchiveTitle, called by PageArchive::listPagesBySearch, which should be accessible from Special:Undelete [16:00:21] tarrow: you probably don't need it though [16:06:19] hey folks, I just sent an email about offsite activities, please respond by tomorrow - thank you [16:15:34] tarrow: oh, i didn't realize you were asking about it still creating the index when turned off. Looking at the code, it seems indeed we don't check like we should. Instead the check is on isPrivateCluster. Basically if you set `$wgCirrusSearchPrivateClusters = []` then it will stop creating them [16:26:00] back [17:03:03] unmeeting, anyone? [17:17:38] ebernhardson: fantastic, that should save us 1/3 indices. Does it have any other impacts? [17:37:04] oh no, don't tell me I didn't get the unmeeting bug [17:37:28] I got the 1x1 invite bug [17:42:38] Well, I hooked gCal up to Slack, maybe that will help. Anyway, eating lunch, back in ~1 hr [17:55:29] Wifi crapped out, will take that as a cue to leave unmeeting anyway I guess [18:46:15] back [19:54:49] Lunch+groceries [20:02:21] hi Search team! would you accept a patch for making CirrusSearch sorts extendable by other extensions (via a hook or configuration + ObjectFactory)? [20:03:21] we use CirrusSearch in GrowthExperiments for surfacing tasks to new users, and want to prioritize between those tasks in certain ways [20:04:18] there are other ways to get that but it seemed like a good thing to have in CirrusSearch [20:08:48] (T301096 is the task where it would be used) [20:08:49] T301096: Add a link: prioritize suggestions of underlinked articles - https://phabricator.wikimedia.org/T301096 [20:28:42] BRB, picking up my kid [20:31:13] tarrow: only change should be that Special:Undelete is harder to use [20:31:30] right [20:31:46] ebernhardson: cool! That's probably a compromise we can take [20:32:04] tgr: my first thought, without looking super deeply into it, is that rescore profiles should allow doing basically this, only difference is they aren't selectable from the 'sort' parameter? [20:41:41] tgr: i suppose what i'm thinking of is a 'script' rescore function chain that implements the desired scoring equation, and a rescore profile that references the function chain. Oddly there are no examples of a profile using the script score function score builder [20:43:43] i guess though i'm baking in the assumption that happens in a rescore instead of as part of the primary scoring, maybe it would be fine to do it as a top level sort [20:50:21] back [21:29:24] thanks! I don't know much about rescoring, I'll read up on it [21:35:46] tgr: the main difference with rescoring is it's a two-stage operation. The primary query runs and selects some X of documents, and then the rescore query can go back in and do expensive calculations only on that subset of documents [21:36:40] its the default way cirrus does things, but elasticsearch has made their direct-sort functionality much more efficient over time which might make it reasonable to use directly