[02:11:24] We really need to work out what to do about old changesets in gerrit... [02:11:54] There's stuff from 2013-2014 that it's obvious no one cares about... [02:15:18] https://gerrit.wikimedia.org/r/#/q/status:open,3900 [12:53:36] legoktm: fyi, I was attempting to change coverme to link to the class in the coverage report, but it seems that's a bit too ambitious in the current approach given core's autoloader complexity. [12:53:42] But maybe that can be easier in the future? [12:54:01] e.g. have ResourceLoader::respond() linked to https://doc.wikimedia.org/cover/mediawiki-core/includes/resourceloader/ResourceLoader.php.html [14:14:22] google fonts cdn - https://twitter.com/matti_sg/status/1001010512786874370 [14:18:53] Reedy: it's almost like we've been right the whole time :) [14:41:48] hey tgr! [14:42:07] hey [14:42:57] i'm trying to figure out how we'll run selenium tests for extensions on mcr-full [14:42:59] I got stuck trying to get the browser tests working on mcr-full [14:43:07] npm, more specifically [14:43:14] ah :) [14:43:24] which somehow refuses to install dependencies [14:43:24] i know very little about that stuff [14:43:36] maybe addshore can help? [14:43:59] same OS as mcr-base, same npm version, same mediawiki checkout, one works, the other doesn't :/ [14:43:59] * addshore is deploying right now but will check back in a bit [14:44:21] huh [14:44:22] I'm not familiar with npm so if someone can help out that would be great [14:45:10] addshore: i'm told we are supposed to have your full attention for mcr starting next week :) [14:45:24] maybe I should just rsync the node_modules dir from the other machine [14:45:31] tgr: in the mean time, how can i make myself useful for mcr-sdc? [14:47:50] I tried to set up the wikidata role on that box which didn't go well, and since we are going to go with a single wiki instead and no valuable work went into that box yet, it should probably be destroyed and started from scratch [14:48:30] ssh: connect to host mcr-sdc.wmflabs.org port 22: Connection timed out [14:48:30] with the wikibase_repo role instead of wikidata [14:48:38] yea [14:48:56] you could do that if you want [14:49:12] mcr-sdc.wmflabs.org is the web proxy, the box is mcr-sdc.mcr-dev.eqiad.wmflabs [14:49:15] i'll give it a try. i havn't done this in years :) [14:49:29] did you use the wikitech ui to select the roles when creating the instance? [14:50:00] yes, silly me - wrong host name [14:50:04] you can destroy/recreate it at horizon.wikimedia.org, wikitech has lost most of its functionality these days [14:50:45] smallest instance type, https://wikitech.wikimedia.org/wiki/Help:MediaWiki-Vagrant_in_Cloud_VPS has a decent step-by-step tutorial [14:52:36] thanks [14:52:42] i'll just copy what i see in the other instances [14:57:29] Permission denied (publickey) [14:59:09] ah, i was too imaptient [15:04:12] tgr: so, in the end, i have to define a new vagrant role for mediainfo and activate it, right? [15:04:22] yeah [15:04:36] similar to https://phabricator.wikimedia.org/source/mediawiki-vagrant/browse/master/puppet/modules/role/manifests/wikibase_repo.pp [15:04:52] just something like mediawiki::extension{ 'WikiBaseMediaInfo': } [15:05:13] where do i put that, and do i do it from inside vagrant, or from the outside? [15:06:00] puppet/modules/role/manifests [15:06:15] vagrant XXX commands only work outside [15:07:10] where you do the file editing only matters to the extent of which user will own it, doing that outside as well is probably less trouble as the repo is owned by the mwvagrant user [15:09:16] vagrant up is taking forever... [15:10:34] If it's from scratch... it will take a while [15:10:50] yeah, that's half an hour or so [15:11:04] maybe less on LXC but still a long time [15:11:09] oh ok. i'll take a walk, then ;) [15:35:26] loading the main page times out... http://mcr-sdc.wmflabs.org/wiki/Main_Page [15:36:26] proxy is up... maybe i didn't configure it right? [15:39:28] maybe the security group is missing? [15:40:23] yeah, it should have the 'vagrant' security group [15:47:48] "it"? [15:48:08] i thought security groups apply to projects? [15:49:22] DanielK_WMDE: they are group by project but applied at an instance level. There is a 'default' (I think?) group that everything in the project gets, but other named groups have to be applied instance by instance [15:50:09] ok. trying to find that in the ui. manual sais to Navigate to "Manage Security Groups". No idea how I'd do that [15:52:15] DanielK_WMDE: yeah, Horizon UI is terrible [15:52:26] it's in the lefthand Network menu [15:52:33] i see the security rules, but i don't see how they are associated with instances [15:52:46] ...no its not, because you want to apply it, not define it, sorry [15:52:58] o_O [15:53:22] Compute > Instances, drop-down menu on the right, edit security groups [15:53:28] yea, just found it. [15:53:34] click on an instance and then "Edit Security Groups" should be in the weird drop down menu on the far right hand side [15:53:40] these days you can even do that without having to rebuild the box! [15:53:41] this is not what the help page says. i'll update it [15:53:58] and yes, their UI is very ... developer centric :/ [15:54:43] if it was developer centric it would have a command line ;) [15:59:24] yay, it's up! http://mcr-sdc.wmflabs.org/wiki/Main_Page [16:00:55] DanielK_WMDE: it does! we just don't give you access to it. There are cli scripts and web apis to do everything [16:03:01] \o/ [16:04:03] bd808: have you seen T196082 before, maybe? it's driving me nuts [16:04:04] T196082: npm install for selenium tests not working on labs-vagrant - https://phabricator.wikimedia.org/T196082 [16:12:13] what am I doing wrong? https://etherpad.wikimedia.org/p/role-mediainfo [16:14:34] DanielK_WMDE: use include instead of require [16:14:43] require blows up 90% of the time [16:15:20] and trying to understand exactly how depenency cycles happen in puppet is a hopeless pursuit IMO [16:15:45] include loads the role but does not create a dependency (ie. runtime ordering) [16:16:27] if one is needed you can add it manually for subresources (e.g. maintenance scripts) [16:16:56] seems to work, thanks [16:17:23] but resources are ordered on the type level (e.g. settings before maintenance scripts) and usually that's enough [16:19:40] hm, now i need to enable file uploads on the wiki [16:20:00] i suppose i'd use mediawiki::settings for that [16:26:37] DanielK_WMDE: there's a role for that I think [16:27:42] role::multimedia [16:29:35] that doesn't set wgEnableUploads it seems... but I can try. setting that from my role didn't work [16:29:38] anyway, time for the call [17:02:33] DanielK_WMDE: wgEnableUploads is true by default [17:02:38] (vagrant default, that is) [17:02:55] /vagrant/LocalSettings.php [17:04:43] DanielK_WMDE: here now, and dam, i missed the mcr call.... [17:04:50] mcr is'nt in my mind again yet [17:04:59] *reads up* [17:05:12] Krinkle: ah hmm, I know how we can do that. The CloverXml object keeps track of file names [17:05:23] legoktm: Woot. [17:05:24] Krinkle: I'll work on a patch [17:05:38] DanielK_WMDE: I didn't think it was full attention, (afaik), but yes, some [17:05:57] legoktm: btw, https://phabricator.wikimedia.org/D1062 - not sure if I used the right method of subscribing you. Did you get a notification from this? (if not, what should I do instead?) [17:06:10] also, should I add an arcconfig file? [17:06:35] Krinkle: I did yes. If that makes it easier for you, sure. I just download the git patches and apply them [17:07:12] legoktm: right, but to submit them to Phabricator, it needs arccconfig, or else 'arc diff' doesn't work. [17:07:21] Or at least, requires parameters that aren't obvious [17:07:36] submit, as in, sharing the patch with you [17:07:48] not the downloading part :) [17:31:02] Krinkle: I'll add you as a maintainer and give you commit access after I finish my meeting [17:49:43] _joe_: Hm.. Ive attempted to move this variable to common.yaml, but it seems it is not being recognised by deployment-prep. Is it not allowed to have :: in that file? I would expect that given the expanding is not enabled in beta (yet) that therefore foo::bar can be defined in common.yaml as fallback. [18:23:27] is $wgDBmysql5 still experimental? [18:42:16] legoktm: What it is is unused. mysql5/tables.sql doesn't exist and 'dbmysql5' and the var are never read anywhere [18:42:27] legoktm: I don't know. As far as I can tell the only thing it does is make DatabaseMysql set the client and/or connection character sets to utf8 (not utf8mb4). I don't know what exactly that does. [18:42:59] Krinkle: MWLBFactory reads it to pass to DatabaseMysql as 'utf8Mode'. [18:43:41] Ah, missed wg-unprefixed variant in my search [18:44:25] INteresting [18:44:34] So that's used to use utf8 instead of binary as connection charset [18:44:43] In DatabaseMysqlBase::connectInitCharset [18:45:07] I'd say just kill it. But /cc AaronSchulz in case he knows more. [19:20:20] at minimum I guess we can have the installer stop writing it in LocalSettings [19:32:47] <_joe_> Krinkle: it should, yes [19:33:18] _joe_: Thanks, found it now. The error about [] being illegal on Nil confused me thinking that the variable was undefined, but it was actually about a different hiera key. [19:33:34] <_joe_> oh, ok :P [19:33:36] _joe_: https://gerrit.wikimedia.org/r/#/c/436581/ - lvs_server [19:33:40] Not sure what to do about it [19:34:24] <_joe_> Krinkle: I'm not completely sure it's a great idea to unify those two lists [19:34:42] <_joe_> people want to deploy to deployment-prep before deploying to production, I would guess [19:35:36] _joe_: Yeah, but I suppose that 1) It can be merged/extended still, and 2) The process we follow for most other things (ini settings, puppet manifest etc.) is to patch prod in puppet, and cherry-pick the patch. [19:35:45] Unless it's specific to beta permanently. [19:36:06] <_joe_> but, I'll take a look at that patch tomorrow, I remember writing that code (scap_source) but that's about it :P [19:36:20] Definining the git repo on tin is harmless though, even if it does have to be done in both. It reflects what we do with mediawiki etensions now. [19:36:34] We no longer support adding an mw extension clone to beta only. We always clone in both. [19:36:35] <_joe_> ack makes sense [19:36:46] but include conditonally. [19:36:59] Thx, I'll see tomorrow then. [19:42:59] James_F: https://gerrit.wikimedia.org/r/#/c/405374/ ;) [21:21:15] legoktm: Just to confirm, dropping the utf-8 option from the installer and its _MysqlCharset setting do not (directly) relate to $wgDBmysql5, right? [21:21:30] Krinkle: nope [21:21:33] The installer option is for _MysqlCharset which is output as part of $wgDBTableOptions [21:21:42] Whereas wgDBmysql5 is a separate variable. [21:21:47] right [21:21:57] Albeit seemingly related, in that wgDBmysql5 triggers connection-time charset options [21:22:19] as opposed to options defined as part of the storage schema. [21:22:21] and it gets written to LocalSettings.php, though I think always default false [21:22:56] "Use binary" in installer results in _MysqlCharset > wgDBTableOptions > substituted in tables.sql [21:23:43] as DEFAULT CHARSET=.. for the tables themselves. [21:24:15] wgDBmysql5 > DatabaseMysqlBase::connectInitCharset -> .. -> SET NAMES ... at query time. [21:24:18] Anyway. [21:24:21] * legoktm nods [21:24:36] I think the latter is easier to get rid of given it's not exposed, marked experimental, and never used in prod or anywhere else notable as far as I know. [21:24:45] and either way, degrades fine, given it's just at query time. [21:31:19] ok, so I'll put up two patches in a bit, one to drop $wgDBmysql5 stuff from the installer (and backport to 1.31), and then another to drop it from core entirely for 1.32? [22:47:09] here's a fun mystery: between 2018-03-04 and 2018-03-05 the number of calls to the Action API per day recorded in hadoop dropped by 2 million and has stayed consistently ~2M lower per day than the prior year. [22:48:02] I see a parsoid deploy on that day, but I don't see any change in it that jumps out as a cause. [22:48:31] * bd808 should run the numbers to see if the 2M can be attributed to internal or external clients [22:49:53] err.. that's 200M, not 2M [22:50:15] dropped by 200 million?? [22:50:36] basically api traffic via this measurement was cut in half starting on 2018-03-05 [22:52:47] the parsoid deploy looks like it was just https://phabricator.wikimedia.org/T188591 [22:52:48] in March of 2017 we were counting around 500M hits per day. in March 2018 that dropped to around 300M hits per day [22:54:19] There's https://www.mediawiki.org/wiki/Parsoid/Deployments#Monday,_Mar._5,_2018_around_1:25_pm_PT:_Y_Deployed_d115592 for the deployment list. [22:54:54] can we see if any user agents disappeared? [22:55:53] "Suppress bogus warning about "variant" parameter to action=parse" https://gerrit.wikimedia.org/r/#/c/415358/ maybe? [22:55:55] * James_F shrugs. [22:56:14] probably... it will take some digging. I'll try the internal vs cloud vs external report first to see if the shift is in an obvious bucket of traffic [22:56:32] * bd808 tries to remember how to hadoop [22:56:55] James_F: that one is just parsoid not logging one of the API warnings to its own logs [22:57:26] Oh right. [23:50:23] Anyone bothered if I stop ObjectFactory testing on PHP55? https://gerrit.wikimedia.org/r/#/c/436701/ so we can use PHP56 features? https://gerrit.wikimedia.org/r/#/c/436699/ [23:52:11] Reedy: fine with me. The splat operator should make things better [23:52:42] that hack I put in was horrible, but soooo much faster than ReflectionClass [23:52:51] heh