[09:01:54] <gehel>	 pfischer: 1:1 ? https://meet.google.com/vkf-mkgd-ywo
[09:15:04] <pfischer>	 gehel: sorry, 2'
[09:17:49] <gehel>	 I'll still be there
[09:58:58] <dcausse>	 lunch+errand
[11:52:03] <gehel>	 dcausse, dr0ptp4kt : I won't be there for the WMF/WMDE sync (conflicting Q2 planning meeting)
[12:16:32] <dr0ptp4kt>	 👀
[12:26:21] <dcausse>	 gehel: np
[13:03:17] <dcausse>	 sigh... I messed the kafka SSL support I think, need to ship a fix soon
[13:04:31] <dcausse>	 wcqs is failing, wdqs will soon fail the first time it seems to send an error to a side-output
[13:10:08] <dcausse>	 might be safer to revert while we review/merge/release a fix
[13:15:20] <dcausse>	 this is the fix https://gerrit.wikimedia.org/r/c/wikidata/query/rdf/+/1075562
[13:21:20] <inflatador>	 <o/
[13:21:29] <dcausse>	 o/
[13:23:25] <inflatador>	 If I can do anything to help w/the wcqs/wdqs stuff LMK
[13:24:00] <dcausse>	 inflatador: am deploying https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/1075561 which reverts back to using plaintext 
[13:24:41] <dcausse>	 should fix the oom that's caussing wcqs to crash
[13:25:25] <inflatador>	 Cool, let me know how it goes...I'll ack alerts
[13:28:32] <dcausse>	 what's weird is that the latency alerts all mentioned wcqs@codfw but I'm sure wcqs@eqiad was affected too...
[13:40:32] <dcausse>	 alert definition looks fine...
[13:40:43] <dcausse>	 anyways wcqs is recovering
[13:43:20] <dcausse>	 no wed meeting today?
[13:52:14] <inflatador>	 dcausse there is an Airflow mtg at the normal time
[14:00:22] <ebernhardson>	 will head out to conf in 30 min, but some random bits so far: many of osc's customers (which are generaly but not exclusively on the more advanced side of search) have transitioned from elastic to opensearch. 
[14:00:56] <ebernhardson>	 they intend to keep maintaining the LTR plugin for elastic, the LTR plugin for opensearch is in a bad space currently (also not available for 1.x iiuc) but they are working on it (after being hired by amazon to do so)
[14:02:12] <ebernhardson>	 amazon does have a person in charge of traditional search, so there is at least some effort invested there, but it sounds like they are currently focused on "user behaviour insights" which helps to collect the data people need for ltr and other boosting techniques
[14:06:44] <ebernhardson>	 other thoughts...there are clearly people using opensearch as a vector database and not just vectors to support search, so that wouldn't be a unique use case if we went that route
[14:38:17] <ebernhardson>	 :q
[15:19:59] <pfischer>	 ebernhardson: Thank you for the summary, sounds promising already.
[15:56:18] <inflatador>	 workout, back in ~40
[16:55:09] <inflatador>	 back
[16:56:15] <ebernhardson>	 my summary of opensearch k8s operator - storage is your problem
[16:57:01] <dcausse>	 :)
[17:43:59] <dcausse>	 dinner
[18:04:11] <inflatador>	 lunch, back in ~40
[18:28:16] <ebernhardson>	 user behavior insights actually seems pretty nifty. We would probably have to sample or otherwise sub-select data that goes into it, but might be interesting
[21:07:59] <inflatador>	 ryankemper in pairing if you wanna join
[21:28:14] <inflatador>	 very much a WIP, but we have a way to track memory usage per wdqs instance now https://grafana.wikimedia.org/goto/oPGxvURHR?orgId=1