[10:18:55] elukey: sorry for writing novels in your patch reviews :D [10:21:57] klausman: nono the opposite, more comments are good :) [10:23:47] I am waiting others to comment for the golang image since it is more complex that I anticipated, not sure if the os-split dirs still make sense or not [10:23:55] we'll see what SRE prefers [10:24:08] I am trying to keep as much compatibility as what they have done so far [10:28:21] Aye [10:28:52] a) don't break stuff if you don't have to b) keep it simple c) make Lift Wing work, is the order of priorities, I'd wager :) [10:34:58] 10Jade: Replace Content::getNativeData() calls with TextContent::getText() in Jade - https://phabricator.wikimedia.org/T283676 (10Aklapper) [10:35:05] 10Jade: Replace Content::getNativeData() calls with TextContent::getText() in Jade - https://phabricator.wikimedia.org/T283676 (10Aklapper) p:05Triage→03Lowest [11:42:19] 10Lift-Wing, 10Machine-Learning-Team, 10Patch-For-Review: Install Istio on ml-serve cluster - https://phabricator.wikimedia.org/T278192 (10Theofpa) > I understand that we'll not have envoy proxy related metrics, but I'd expect to see some metrics about how istiod works (pilot metrics maybe) and possibly also... [11:42:54] elukey: I think Guiseppe's proposal re: image names for the Go image is a great idea. [11:57:20] 10Lift-Wing, 10Machine-Learning-Team, 10artificial-intelligence, 10revscoring: Create generic revscoring inference service - https://phabricator.wikimedia.org/T283526 (10kevinbazira) **>>FILES<<** On the KFv1.1 sandbox, I have been able to create a generic revscoring inference service using the files belo... [13:28:17] klausman: yes it makes sense, will update the code review in sec [16:26:13] klausman: thanks for the additional code review on the inference-services, it's very nice to have more eyes on stuff [16:27:42] accraze: o/ [16:27:46] in the past it was mostly just me & kevin lol forgot what it's like to have more team members [16:28:03] one question about STORAGE_URI: does it need more sidecar/configs/horrors to kfservign? [16:28:15] elukey: hey hey [16:29:26] good question, both kevinbazira & I are pulling images directly from a public s3 bucket right now with no additional config outside of setting an env var [16:30:09] the only other config we might need is a secret for thanos swift. i have a WIP patchset for that here: [16:30:11] https://gerrit.wikimedia.org/r/c/machinelearning/liftwing/inference-services/+/693217 [16:30:28] accraze: we have all code reviews lined up for our images in https://gerrit.wikimedia.org/r/c/operations/docker-images/production-images/+/695173 (there is a chain of them) [16:30:43] as soon as we have them on our docker registry we should start using them if possible [16:30:54] ^ whoa this is awesome!! [16:30:55] I'll also try to add istio 1.10 [16:31:48] yeah i think the next thing i really want to do is run is to try running our services on all of our own images [16:32:24] lol need more coffee [16:33:33] super [16:43:30] elukey: is there any documentation about how to make a new base image and push to our registry? [16:45:21] also i did some experimentation with external transformers yesterday for the outlinks topic model [16:45:48] might have run into a TLS issue with the local gateway, but plan on digging into it a bit more today [16:45:49] ah yes I wanted to ask you about the transformers.. are those separate containers? [16:46:42] yep -- separate containers. I have a patchset up: https://gerrit.wikimedia.org/r/c/machinelearning/liftwing/inference-services/+/694756 [16:47:36] it seems to work when i use the public ingress, but when I use the local cluster address it runs into trouble [16:48:31] ah it is a custom image from you okok [16:48:53] with local cluster address you mean the cluster local gateway? [16:49:04] ^ yeah that's the one [16:49:08] (ETOOMANY names in my head sorry) [16:49:14] :D [16:49:16] LOL that was me yesterday [16:49:42] for the base images, I am not sure if we have some docs [16:49:48] but we use a tool called docker-pkg [16:50:01] and the production-images gerrit repo [16:50:12] it is basically a nice wrapper for docker build etc.. [16:51:03] for the moment let's use your dockerhub account for testing, once the istio/knative/etc.. images are code reviewed and built we can look into that [16:51:08] does it work?? [16:51:41] perfect! i think kevinbazira is using his account too for now [16:52:17] once we know the other images are good to go, then we can start talking about how to handle our custom images in production [16:54:29] i'll just say, injecting the model binaries from storage into a serving container has been really nice so far [17:08:24] 10artificial-intelligence: Qualitative data collection for ores bias analysis - https://phabricator.wikimedia.org/T225441 (10Groceryheist) 05Open→03Resolved [17:08:26] 10ORES, 10artificial-intelligence, 10editquality-modeling, 10Machine-Learning-Team (Active Tasks): Find out what tools are used for making reverts on the ores-enabled wikis. - https://phabricator.wikimedia.org/T225134 (10Groceryheist) [17:10:29] 10Machine-Learning-Team, 10MediaWiki-extensions-ORES, 10ORES, 10Growth-Team, 10Growth-Team-Filtering: Why are there three Q-marks (???) in threshholds in Special:ORESModels? - https://phabricator.wikimedia.org/T221870 (10Groceryheist) 05Open→03Stalled [17:10:58] 10Machine-Learning-Team, 10MediaWiki-extensions-ORES, 10ORES, 10Growth-Team, 10Growth-Team-Filtering: Why are there three Q-marks (???) in threshholds in Special:ORESModels? - https://phabricator.wikimedia.org/T221870 (10Groceryheist) 05Stalled→03Resolved a:03Groceryheist [17:11:00] 10ORES, 10Research ideas: Analyze the effects of ORES deployments on counter-vandalism behavior - https://phabricator.wikimedia.org/T200898 (10Groceryheist) 05Open→03Resolved [17:11:54] 10ORES, 10artificial-intelligence, 10editquality-modeling, 10Epic, 10Machine-Learning-Team (Active Tasks): ORES bias analysis - https://phabricator.wikimedia.org/T224901 (10Groceryheist) [17:11:59] 10ORES, 10artificial-intelligence, 10editquality-modeling: Fit models for revert prediction - https://phabricator.wikimedia.org/T224902 (10Groceryheist) 05Open→03Declined This project ended up going a different direction. [17:23:03] accraze: one thing that it will probably be needed is a k8s egress rule to fetch the model from swift (nothing big but let's remember that) [17:23:48] 10Machine-Learning-Team, 10SRE, 10Release-Engineering-Team (Radar): Contact number of some WMDE staff should be avalible to SRE/RelEng - https://phabricator.wikimedia.org/T210721 (10WMDE-leszek) 05Open→03Resolved wikidata-emergency@wikimedia.de email address is meant to be used as a means of reaching WMD... [17:24:03] accraze: ping me any time for code reviews if you like more eyes on things [17:52:34] * elukey afk! [18:13:03] 10Lift-Wing, 10Machine-Learning-Team, 10artificial-intelligence, 10revscoring: Create generic revscoring inference service - https://phabricator.wikimedia.org/T283526 (10ACraze) @kevinbazira this is great news! glad to hear the generic container approach is working so far. I have uploaded the enwiki-damagi... [18:42:05] 10Lift-Wing, 10artificial-intelligence, 10Machine-Learning-Team (Active Tasks): Create a tutorial for deploying a model on toolforge - https://phabricator.wikimedia.org/T281317 (10Isaac) FYI -- Research has two templates for this that might be of use: * API on Cloud VPS: https://github.com/wikimedia/research...