[11:32:55] lunch [14:15:56] o/ [14:29:49] \o [14:32:24] o/ [14:33:02] analysis-icu plugin is now in our OS image: https://gitlab.wikimedia.org/repos/data-engineering/opensearch . Working on opensearch-innerhits now [14:33:16] thanks! [14:33:42] inflatador: nice! i have released a new version of that plugin yesterday, it's now -wmf6 [14:34:16] np. I know the blubberfile is pretty gross, so if y'all have suggestions LMK. I figure we will eventually release a deb pkg with all the plugins eventually [14:34:44] we might update it again though, but getting the process in place is probabably worth it. There is the opportunity to do some of what i was going to do in the ml connetor pre_process_function, but it feels so awkward building json with a StringBuilder that i was trying to avoid it [14:34:59] s/ml connnector/inner hits processor/ [14:35:07] err, awkward wording all around :P [14:40:08] ended up installing java on relforge and mounting: /etc/ssl/certs/java/cacerts:/etc/ssl/certs/java/cacerts [14:40:48] added -ub 8196 to llama otherwise it fails for me [14:43:47] dcausse: awesome, so starting up the opensearch container from docker-compose should just work now? [14:43:59] yes that's my hope? :) [14:44:26] haven't tested a restart yet because I'm extracting query results but will do a quick test right after [14:53:20] done two weekly runs of spark-nlp to compute the delta of new/modified passages for en, fr and ptwiki and seems to take less than 3hours with 500cores [14:54:38] snice! [14:57:57] ebernhardson: btw if you have a moment could you take a look at https://gitlab.wikimedia.org/repos/search-platform/discolytics/-/merge_requests/73? I'd like to setup the job in airflow [14:58:10] dcausse: sure [14:58:14] thanks! [14:58:15] ah, we have a bit of a glitch with the plugin. It supports opensearch 3.5.0 but the deb image is on 3.3.2 [14:59:02] I can try syncing our OpenSearch repo to upstream, maybe we'll get 3.5.0 [14:59:30] inflatador: oh, i hadn't thought of that. I can almost certainly build it for 3.3.2, 3.5.0 was just the latest [15:00:15] ebernhardson yeah, go ahead and build if it's not too hard. I'll also look at syncing the repo regardless [15:00:26] it might just be changing a number and running a build, testing now [15:04:22] let me know how it goes, looks like I can also update the deb pkg: `Updates needed for 'trixie-wikimedia|thirdparty/opensearch3|amd64': [15:04:22] 'opensearch': '3.3.2' will be upgraded to '3.5.0' (from 'thirdparty/opensearch3'): [15:04:22] files needed: pool/thirdparty/opensearch3/o/opensearch/opensearch_3.5.0_amd64.deb` [15:17:57] minor adjustments to jackson versions, but otherwise looks like it should work. Will push a patch and trigger a new build in a minute [15:19:58] cool, if that doesn't work LMK and I'll try and work on a 3.5.0 image. [15:34:21] hmm, :forbiddenApisMain passes locally but fails in CI :S [15:42:59] ahh, i configured CI to use the java25 image, but the forbiddenApi version 3.3.2 depends on doesn't have the specs for that, but 3.5.0 does. [15:43:26] i'm using 17 locally, and we have a java17 image, lets see.. [15:44:12] err, hmm i guess i've been using 21. Will see anyways :) [15:44:18] yeah, should be 21 [15:44:32] the thing is we dont have a java21 image at https://docker-registry.wikimedia.org/ [15:44:51] looks like 8, 11, 17, 25 [15:45:34] yea compat tables say 3.2.0+ needs 21, hmm [15:47:00] maybe i can just tell forbiddenApis to use an older java spec [15:48:23] we do have a java 21 image, it's based on trixie though [15:48:40] inflatador: hmm, which one? I was looking at the javaNN images [15:48:46] docker-registry.wikimedia.org/openjdk-21-jre:0.2 [15:48:54] ahh, ok lemme try that [15:48:57] maybe there's not a JDK one? I can make one if there isn't [15:49:12] oh, yea if it's just JRE it might not work [15:49:48] nah, looks like there is a jdk version https://docker-registry.wikimedia.org/openjdk-21-jdk/tags/ [15:58:20] quick errand [15:59:53] mildly annoying, the openjdk-21-jdk image defaults to the root user, but the opensearch test cluster refuses to run as root. And it doesn't look like .gitlab-ci.yml has a directive for that. Can work arround it [16:13:12] looks to have worked, building now [16:57:28] (trigger_release is being annoying :P but getting there) [17:00:31] workout, back in ~40 [17:05:54] inflatador: this should work: https://gitlab.wikimedia.org/repos/search-platform/opensearch-innerhits/-/packages/1964 [17:15:19] heading out, have a nice week-end [18:08:43] OK, back for real [18:18:22] build job is running https://gitlab.wikimedia.org/repos/data-engineering/opensearch/-/jobs/746306 [18:35:23] OK, looks like we have an image with the innerhits plugin , `docker pull docker-registry.wikimedia.org/repos/data-engineering/opensearch:3` if you wanna give it a spin [18:35:30] thanks! [18:36:18] oops [18:36:22] I never actually merged it ;( [18:36:24] one sec [18:37:37] OK, merged for real now