[13:40:15] wiki-ai/revscoring#593 (flake8 - 0820ea0 : amir): The build passed. https://travis-ci.org/wiki-ai/revscoring/builds/116901025 [14:55:41] halfak: o/ [15:37:27] Amir1, o/ [15:37:40] Sorry for no assignments. Was out late last night. [15:37:57] np, I found some stuff to do [15:38:02] Great. [15:38:02] not very serious things [15:38:06] but fun ones [15:38:08] Docs and polish [15:38:09] :) [15:39:06] I'm testing my updates to requirements.txt in ores-wikimedia-config right now. [15:39:20] Regretfully, I have to let a bunch of crap compile :( [15:39:27] But I think I'm almost done. [15:39:34] great [15:39:46] I also made my first differential patch [15:39:47] Our new wheel strategy is going to make a few things easier with our deploymentsa. [15:39:47] https://phabricator.wikimedia.org/D154 [15:40:03] It looks nice comparing to gerrit [15:40:16] Cool! Yeah. [15:40:23] Would like to start working with this. [15:40:38] Does it use the gerrit style patchset pattern? [15:40:52] I don't think so [15:40:58] it's more like github [15:41:29] you can make several patches and make a differential [15:44:24] Interesting. [15:44:34] Can you fork a repo? [15:46:23] I'm not sure [15:46:37] I haven't seen anything likewise [17:00:27] halfak: I made a PR in revscoring. Can you check that out? [17:00:32] it's pretty simple [17:40:09] Hey Amir1 [17:40:13] looking now [17:40:27] awesome [17:40:28] thanks [17:41:30] {{merged}} [17:41:31] 10[1] 04https://meta.wikimedia.org/wiki/Template:merged [17:42:31] \o/ [18:04:32] halfak: got a quick question for you. Just making sure of something. Are requests to ORES straight from the user's browser/app ? or do they come from a mediawiki extension ? [18:06:15] akosiaris: both [18:07:08] hmm ok, actually the former is a superset of the latter so that answer my question. [18:07:12] Amir1: thanks! [18:07:25] yw, I'd be happy to help [18:07:54] :) [18:23:48] Thanks Amir1 [18:23:53] was distracted IRL [18:29:26] :) [19:11:35] Amir1, just doing my final tests of models before pushing changes to ores-wikimedia-config [19:11:49] Can you help me find a few vandalism edits in arwiki? [19:11:58] awesome [19:12:03] have you updated wheels? [19:12:18] I might not be very good at it [19:12:45] but if you send me some reverted edits I can check [19:13:15] Oh yeah! Forgot that I can just look at our train/test set [19:13:25] I just finished updating the freeze list [19:13:39] Will be pushing that to a PR and then I'll get started on generating wheels. [19:14:50] For some reason, we're only getting a single decimal place for our probabilities with the arwiki model. [19:16:45] But it seems to point at the reverted edits more than the non-reverted edits. [19:16:55] I'll get it on staging and then have you take a look. [19:17:00] hopefully :) [19:18:10] sure [19:21:04] halfak: you need to update wheels repo and update it in ores-wikimedia-config (as a submodule) [19:21:36] Amir1, yup [19:21:40] Hence WIP :) [19:22:10] cool [19:22:34] I'm working on ores-compute-01 now to get the wheels. [19:22:41] We should add a puppet role for "compute" [19:27:50] halfak: in the mean time can you check https://phabricator.wikimedia.org/T130369 [19:29:34] Amir1, wondering if we should solve parts of the process before engineering it all. [19:29:45] E.g. setting up CI that will deploy current master to staging. [19:31:23] yeah I want us to discuss about the best approach and then we start working [19:32:00] kk. Would be great to have releng in on this. Want to go bring it up in that channel? [19:32:43] hmm [19:32:54] I think hashar would be the best person [19:32:58] Agreed. [19:33:10] he wanted to make ores service working in beta clsuter [19:33:30] Oh. So we'd set up our staging environment in there somehow? [19:36:52] we can have several test systems, one would be beta [19:36:59] ores beta [19:37:11] not staging [19:37:24] halfak: ^ [19:37:32] Sure. Would like to get a sense for what it means to "be in beta cluster" [19:38:22] the whole thing is we send everything merged directly to ores beta [19:38:36] so if anything breaks we would know [19:45:25] Amir1, yeah. I think I get that. [19:45:42] Just -- what exactly is the "beta cluster" and how do we "get in it"? [19:46:21] AFAIK it's a cluster similar to production cluster [19:47:29] we should ask hashar for more info [19:47:48] kk [19:47:53] Looks like he might be AFK [19:48:15] Amir1, I propose we rename your task to "Setup ORES in Beta cluster" [19:48:28] We could also just create a new task and let yours be the [Epic] [19:49:43] scipy is stupid [19:49:59] I'm really upset that they have numpy as a dependency in their setup.py [19:50:04] What a dumb thing [19:50:32] halfak: I think this should be an epic task [19:50:33] So, you can't know that scipy requires numpy until it is too late and you have an import error [19:50:40] Amir1, cool make it so! [19:50:52] because this only one of part of the whole pipeline [19:50:54] sure [19:51:15] halfak: tell me about it! [19:51:19] been there [19:51:28] it still aches [19:53:02] halfak: what about other parts of the task? [19:53:09] packing the whole thing [19:53:12] Amir1, I'm going to kill ores-compute from the `revscoring` project [19:53:14] OK With you? [19:53:28] go ahead [19:53:32] :) [19:53:33] Amir1, I'm happy having those be blockers for the Epic task [19:53:43] If we decide against them later, that is fine. [19:54:26] sure [19:57:26] ores-compute is dead. Long live ores-compute-01/02! [19:59:17] \o/ [19:59:37] We have a new project for staging to live in now too [19:59:39] ores-staging [20:00:01] this is sooo cool [20:00:02] mutante insisted that we stop putting our test instances for ores in `revscoring` project [20:00:22] -._o_.- [20:01:23] halfak: btw I can make to this week's meeting but since the university is closed for new year's holiday I might use skype and you know how that goes [20:04:45] Amir1, gotcha. This is 3/21 meeting? [20:04:55] yes [20:05:05] Oh! We moved it to 3/22 [20:05:07] Does that help? [20:05:08] the next would be as the same [20:05:28] no it closed from tomorrow until the next week [20:05:39] (weeks start from Saturday here) [20:06:25] classes start form April 1 [20:06:26] kk [20:06:50] I'll do my best to get everyone on skype. [20:11:42] Such a pain in the butt to get all of the weird dependencies installed for scipy & numpy. [20:21:11] halfak: https://phabricator.wikimedia.org/T130369 [20:21:29] one of subtasks has a subtask [20:22:07] https://phabricator.wikimedia.org/T130411 [20:22:09] I start with this [20:26:33] Amir1, looks good. [20:26:35] halfak: do you want to work on the swagger [20:26:48] we should split to work [20:26:59] Amir1, will pick that up soon, yes. [20:27:14] since I want this to be done and I don't like the task enough to do it on my own [20:27:24] :D [20:44:37] Amir1, git review -R [20:44:39] Woops [20:44:43] https://gerrit.wikimedia.org/r/278351 [20:44:55] ? [20:45:03] yeah [22:01:42] Well.. I think I borked this patchset [22:01:45] But I had it working [22:01:56] I might just abandon the current patchset [22:02:00] And start a new one [22:02:02] Amir1, ^ [22:02:16] Somehow, I managed to get all of the nltk data included. [22:02:40] do you want me to do it halfak ? [22:02:56] Na. I think I got this. Let's see how smart gerrit is :) [22:03:47] pretty dumb I must say :D [22:04:15] * halfak hopes that gerrit reads his commit message to find the old change Id [22:04:24] Heh. Instead it seems to have just locked up [22:04:55] It worked! [22:04:57] https://gerrit.wikimedia.org/r/#/c/278351/ [22:05:04] Amir1, ^ [22:05:22] \o/ [22:05:23] Git is trying to be smart by assuming that files have been renamed. [22:05:27] But I didn't tell it that [22:05:29] let me check [22:05:54] Looks like celery-3.1.23-py2.py3-none-any.whl has a high similarity index [22:05:59] 89% [22:06:04] so it assumed that was a move [22:06:35] halfak: why Levenstien is deleted [22:06:40] I think we need it [22:07:12] I couldn't find any dependency on it and everything seems to work without it. [22:07:44] OK [22:08:01] https://github.com/wiki-ai/wb-vandalism/blob/master/wb_vandalism/features/diff.py#L247 [22:08:15] Looks like we used to measure edit distance for labels and description [22:08:26] But we don't have that feature anymore. [22:09:32] halfak: it's merged [22:09:36] yeah I know [22:10:19] we should add it later [22:11:12] kk [22:12:20] halfak: update wheels submodule in ores-config [22:12:39] (it's still WIP so my guess is you want to do it ) :D [22:12:57] yup. working on it now [22:14:25] Amir1, https://github.com/wiki-ai/ores-wikimedia-config/pull/48 [22:14:27] Ready [22:14:38] Once you merge this, I'll try pushing to staging. [22:15:30] merged [22:18:00] Amir1, I'm going to kill a bunch of old ores-wm-config branches [22:18:05] I'll leave scap alone [22:18:09] cool [22:19:14] SPAM! [22:22:42] Arg! Our setuptools dependency breaks the install [22:24:06] Amir1, https://gerrit.wikimedia.org/r/278406 [22:24:39] I noticed this when testing on ores-compute-01, but forgot to remove it in the commit [22:25:27] oh, I had it too [22:25:32] and I deleted it [22:30:26] ImportError: cannot import name 'libsvm' [22:30:28] wat [22:31:37] Weird. Our whls don't seem to work on staging [22:32:19] I think it might be because staging is old [22:32:43] I'm going to start up staging-02 and run a test [22:34:49] https://gerrit.wikimedia.org/r/278407 [22:36:12] {{done}} [22:36:13] How efficient, halfak! [22:37:29] \o/ [22:43:54] Well that was fun. Just lost internet for 10 mins. [22:44:05] I'm almost done setting up the new staging server [22:50:03] Woops. Looks like we need to update puppet for our new languages [22:50:45] * halfak works on that [22:54:04] oh, yeaaah [22:55:08] https://gerrit.wikimedia.org/r/#/c/278413/ [22:55:09] BTW [22:55:28] I just manually installed them in staging so that we could test [22:55:39] So we're not blocked by that [22:55:57] http://ores-staging.wmflabs.org/ !!!! [22:58:12] Amir1, http://ores-staging.wmflabs.org/v2/scores/enwiki/reverted/4356789/ [22:58:17] We're live with new code! [22:58:25] And you can see that the v2 paths work. [22:58:37] Lots more details: http://ores-staging.wmflabs.org/v2/scores/enwiki/reverted/ [22:58:45] \o/ [22:58:46] \o/ [22:58:57] So, we're very close to pushing this live. [22:59:05] However, I think we're going to need to stand up new VMs. [22:59:14] I had version issues with our old staging VM. [22:59:36] I think that, in order to deploy our wheels, we'll need to stand up a whole new set of web & worker nodes :/ [23:00:54] This shouldn't be a blocker for prod, but it will be a blocker for wmflabs cluster updates :\ [23:01:38] * halfak prepares to kill old staging-01 [23:02:35] OK. that's it for me. [23:02:45] I'm outa here. Have a good one!