[08:53:39] (03Abandoned) 10Hashar: Jenkins job validation (DO NOT SUBMIT) [extensions/ORES] - 10https://gerrit.wikimedia.org/r/436523 (owner: 10Hashar) [10:57:35] 10Scoring-platform-team, 10articlequality-modeling, 10artificial-intelligence: Build a gadget to give list of quality of articles in categories - https://phabricator.wikimedia.org/T202745 (10He7d3r) Maybe this is useful? https://github.com/he7d3r/mw-gadget-ScoredCategories I was using it to check a few categ... [12:06:12] 10Scoring-platform-team (Current), 10ORES: Migrate gadgets that use ores labs instance to prod - https://phabricator.wikimedia.org/T202630 (10Ladsgroup) \o/ ``` ladsgroup@mwmaint1001:~$ mwgrep 'ores.wmflabs.org' --user (total: 0, shown: 0) ladsgroup@mwmaint1001:~$ mwgrep 'ores.wmflabs.org' (total: 0, shown:... [12:15:44] akosiaris: hey, this is not merged :/ https://gerrit.wikimedia.org/r/c/operations/puppet/+/454577 [12:15:50] has +2 though [12:21:38] Amir1: aha, my bad. Sorry for that. Merging now [12:27:26] Thank you ^^ [14:25:20] 10Scoring-platform-team (Current), 10ORES, 10articlequality-modeling, 10User-Ladsgroup, 10artificial-intelligence: Rename wp10 ORES model - https://phabricator.wikimedia.org/T196240 (10Ladsgroup) a:03Ladsgroup I get it done [15:04:17] 10Scoring-platform-team (Current), 10ORES, 10articlequality-modeling, 10User-Ladsgroup, 10artificial-intelligence: Rename wp10 ORES model - https://phabricator.wikimedia.org/T196240 (10Ladsgroup) https://github.com/wikimedia/ores-wmflabs-deploy/pull/98 I merge this and push it to labs. [15:26:10] I may be late to the scoring platform meeting today. [15:38:50] (03PS1) 10Ladsgroup: Rename wp10 to articlequality, make wp10 alias [services/ores/deploy] - 10https://gerrit.wikimedia.org/r/455585 (https://phabricator.wikimedia.org/T196240) [16:02:56] 10Scoring-platform-team, 10ORES: Send error logs to logstash - https://phabricator.wikimedia.org/T168921 (10hoo) a:03hoo [16:10:09] Amir1: I was just told of https://phabricator.wikimedia.org/T202764. It might be related to wikidata itself, thought you might want to know [16:10:38] akosiaris: Thanks [16:15:30] 10Scoring-platform-team (Current), 10Wikilabels: Extend wikilabels to support session-labelling - https://phabricator.wikimedia.org/T201370 (10notconfusing) I loaded a test dataset to https://labels-staging.wmflabs.org/ it was `en` for now. ### Definitions: + two or more edits within registration (could be... [16:18:50] This is awkward, but I don’t seem to be able to log in to my computer. Count me out for now. [16:26:32] And I'm in! [17:14:25] harej: it would be great if you take a look at the notes of the meeting and let me know about anything that you have opinion or idea or concern. Thank you! [17:14:33] afk for lunch/dinner [17:37:17] Who does the primary work for Wikilabels? [17:37:43] awight: Sorry to have missed 90% of the meeting. Do you have any interesting updates on JADE or its architecture? [17:43:39] harej: wikilabels is a halfak-Amir1 collaboration AFAICT [17:44:16] harej: I don't have anything new to report wrt. JADE, but have been going through the RFC meeting notes and dropping summaries into the task, FWIW. [17:45:40] I've decided to include the secondary table schema in this first release, also. [17:46:11] Normally I'd prefer to deploy that in a second phase, but I think we need to provide in order to better ground the discussion. [17:47:25] Had some weird thoughts about MCR, but still chatting with DKinzler about that. [17:48:39] Basically, I'd like to go with his suggestion to hang page judgments off of pages, in a new slot. Ideally, I'd like to see all entities handled this way, but it seems that Page is the only place it'll be appropriate for now. [17:49:30] Another thing on my mind (and now I'm completely contradicting my "nothing new to report" status) is that we might be able to get rid of the polymorphic/generic interfaces and provide a specific API for judging each entity type. [17:50:35] I can't think of any workflows which would deal with more than one entity type, so the generic aspect is just error-prone cruft from the client perspective. [18:11:38] back now [18:11:42] * Amir1 is reading [18:13:20] awight: if the schema is ready, let me know so I review it. I'm rather good with schema designs [18:13:42] thanks to this book mostly: shop.oreilly.com/product/0636920022343.do [18:14:26] :D great to hear, you'll be the first to know. [18:14:49] I'm writing an unrelated mw-core unit test at the moment, but might have a schema by later today. [18:14:52] It won't be much... [18:20:03] awight: it would be great if you review https://gerrit.wikimedia.org/r/c/455585/ because I want to test it on beta cluster and deploy it in the window [18:20:19] Already tested on labs [18:24:55] ok, reviewing [18:26:06] (03CR) 10Awight: [C: 04-1] "typo needs fixing" (031 comment) [services/ores/deploy] - 10https://gerrit.wikimedia.org/r/455585 (https://phabricator.wikimedia.org/T196240) (owner: 10Ladsgroup) [18:27:53] (03PS2) 10Ladsgroup: Rename wp10 to articlequality, make wp10 alias [services/ores/deploy] - 10https://gerrit.wikimedia.org/r/455585 (https://phabricator.wikimedia.org/T196240) [18:28:09] (03CR) 10Ladsgroup: Rename wp10 to articlequality, make wp10 alias (031 comment) [services/ores/deploy] - 10https://gerrit.wikimedia.org/r/455585 (https://phabricator.wikimedia.org/T196240) (owner: 10Ladsgroup) [18:33:35] (03CR) 10Awight: [C: 032] Rename wp10 to articlequality, make wp10 alias (031 comment) [services/ores/deploy] - 10https://gerrit.wikimedia.org/r/455585 (https://phabricator.wikimedia.org/T196240) (owner: 10Ladsgroup) [18:35:05] Amir1: you said there was a PR as well, feel free to send any time! [18:36:19] awight: yup, this one: https://github.com/wikimedia/ores/pull/260 [18:36:25] ty [18:36:47] It doesn't pass IPs but I want to implement it in another PR [18:37:58] Amir1: Where is primary / secondary documented? I see no mentions in https://www.mediawiki.org/wiki/PoolCounter [18:38:42] The extension has https://www.mediawiki.org/wiki/Extension:PoolCounter#Configuration [18:38:45] awight: AFAIK that's what akosiaris said to me, it should be able to fall back to the secondary node if it can't connect to the primary [18:38:47] a list of servers [18:39:03] ah--would you mind implementing as an array, to match other usages? [18:39:06] mediawiki also shards but that's not needed in our case [18:40:10] awight: hmm, that means it will shards the request (we can make it go random), would that work for you? [18:40:26] wait wat [18:40:32] Meaning, round-robin? [18:40:59] ah no I see [18:41:05] sorry I should have read the thing I linked [18:41:41] consistent hashing does seem right, but akosiaris's advice makes the architecture sound like something else. [18:43:12] I was stupid, the sharding based on randomness doesn't work [18:43:38] yah sounds like consistent hashing on IP ftw [18:43:52] (03CR) 10Ladsgroup: [V: 032] Rename wp10 to articlequality, make wp10 alias [services/ores/deploy] - 10https://gerrit.wikimedia.org/r/455585 (https://phabricator.wikimedia.org/T196240) (owner: 10Ladsgroup) [18:44:06] however, if it's really a primary-secondary thing, then the PoolCounter docs are wrong. [18:46:27] Alex said we can use both sharding and fallback, implementing both wouldn't be hard [18:46:50] but we need to determine the secondary and primary anyway [18:47:13] unless we make the assumption that first member of the array is primary [18:47:21] but I don't like it [18:48:24] I don't understand how we can use consistent hashing to a single server. Are there primary and secondary clusters? [18:50:18] no, primary and secondary nodes [18:50:40] so poolcounter1004 and poolcounter1005 or something like that [18:50:52] plus poolcounter2004 and poolcounter2005 [18:52:42] So where does consistent hashing come into play, as mentioned in Extension:PoolCounter... [18:53:32] awight: Extension:PoolCounter is a client like ores will be a client, we can implement consistent hashing or we can ignore and have fallback way of working [18:54:20] given the load on the extension and the load on ores, hashing is not that useful unless we get lots of outside request which I doubt [18:55:36] No need to prematurely optimize, but I do think we should be worried about the additional delay of a mutex like this. [18:55:55] Mostly, I was concerned that we're getting contradictory information [18:58:24] awight: I talked to Tim and he said the response time from the poolcounter is 10 micro seconds and by adding the network overhead it will be less than a millisecond [18:58:33] it's designed to be fast [18:59:22] :) unless the primary node is down and we're waiting for a TCP timeout [18:59:34] programmers always give best-case numbers :p [19:00:28] if the node is down it should give ConnectionRefusedError but in practice everything can happen :D [20:11:58] 10Scoring-platform-team (Current), 10ORES, 10articlequality-modeling, 10Patch-For-Review, and 2 others: Rename wp10 ORES model - https://phabricator.wikimedia.org/T196240 (10Ladsgroup) I tested it and it worked on beta cluster but it failed on canary node in prod majestically: ``` ladsgroup@ores1001:/srv/d... [20:38:57] Amir1: Everything okay? You have me worried ;-) [20:39:17] the deployment? [20:39:27] it's fine :D [20:39:42] I'm writing the announcement now [20:43:19] whew! [20:48:45] 10Scoring-platform-team (Current), 10ORES, 10articlequality-modeling, 10User-Ladsgroup, and 2 others: Rename wp10 ORES model - https://phabricator.wikimedia.org/T196240 (10Ladsgroup) [20:56:07] I'm done for the day, will bring my work laptop to home just in case [20:56:14] o/ [21:03:21] oh god please don't ;-) [21:03:35] have a relaxing midnight! [21:14:01] awight: should I assume JADE supports all of the same things as ORES? [21:14:54] harej: Very different things, so I'm not sure what you mean. [21:15:08] Is this about the bit where JADE judgments are included in ORES scores? [21:15:39] So JADE lets you tag an edit as damaging or not damaging. Does JADE also let me, for instance, mark a draft article as spam or say that a given article is topic XYZ? [21:15:45] These are all things ORES does. [21:17:23] ah I see, yeah the schemas happen to be almost exactly the same for now. [21:17:47] Draft topic is a bit different, since we're letting the topics be freeform rather than enforcing our odd mid-level topics. [21:17:55] itemquality doesn't exist yet. [21:32:18] back in10 [22:50:44] awight: can you send me the list of wikis with ORES? [22:51:04] both the wikis with the ORES extension and wikis that have models developed for them [23:00:00] awight: never mind, found my original list (from the last time I asked you ;) [23:01:35] hehe [23:02:03] yeah the first is wmf-config and the second is ores config [23:43:15] and Max Klein's main work these days is Wikilabels, right? [23:48:49] harej: notconfusing is working on a change to allow labeling of entire edit sessions, IIRC [23:49:08] Hello [23:49:32] yes, there's already an experimenatl version of it you can try out [23:49:36] o/ neat! [23:49:56] It sounded like harej was asking from a PM perspective, not sure. [23:50:32] harej: is there some PM questions I can answer? [23:52:43] notconfusing: would you hypothetically be up for building JADE support into Wikilabels, or is that out of your scope of work? [23:53:07] As far as I know it would mostly be a check of "if this wiki has JADE enabled, perform this write action" [23:55:47] it's peripherally in my scope. I'm not sure what it really entails. (My main focus is to build a bitten-newbie, or promising newbie detector ATM for a civilservant.io experiment) [23:55:56] Aha. [23:56:23] it sounds like it might be easy, and it might make the promising-newbie detector more accurate, so... maybe [23:57:03] To be honest I'm not sure what impact it would have on your model if it's just data storage. [23:57:26] In any case, it'll be a while before we begin thinking about JADE integrations actually happening. [23:58:36] awight: I'm working on a list of wikis based on what software features they have / software tools they support, with the idea that it's worth prioritizing the wikis that (a) have already bought into ORES and (b) support Huggle and Wikilabels, which are currently my two priorities for integrating (I'm open to other ideas)