[00:00:49] urandom: it's pretty handy. A bit easier to manage than just using flags [00:01:05] mostly because I bounce around from project to project a lot [00:02:53] bd808: how does flags work? [00:04:24] bd808: that might not be the right question, what did your flags-based workflow look like? [00:04:27] urandom: check out https://phabricator.wikimedia.org/flag/ [00:04:58] I have stored searches for each color with labels like "Platform deps" and "Labs & Tools" [00:05:23] urandom: you associate a sense of political identity with some graphical representation through repeated exposure and by imbuing the graphic with totemic significance by integrating it with various rituals [00:05:33] and shortcuts on my personal dashboard -- https://phabricator.wikimedia.org/dashboard/view/5/ [00:05:36] ori: ha! [00:05:52] over time you can use the graphic to elicit certain forms of behavior [00:06:01] This has been another episode of "Fun with Flags featuring ori" [00:06:38] I'm indispensable, what can I say [00:06:51] urandom: the thing about flags in phab is that they are private. YOu can't see my flags even if I shared a search with you [00:07:38] ori: https://www.youtube.com/watch?v=LeyofQK6tRw [00:08:06] bd808: ok, so good for self-organizing, but not as good as a workboard for sharing that with others [00:08:13] yeah [00:09:56] bd808: cool; thanks! [00:10:08] oh Joel is getting fancy and making sub projects of his personal project -- https://phabricator.wikimedia.org/project/view/1788/ [00:11:38] huh [01:23:42] * robla totally got nerdsniped by bd808's observation, and started futzing with ArchCom related Phab stuff. [15:28:19] MatmaRex: I could use your opinion on https://phabricator.wikimedia.org/T128054 as to what the best way to fix it would be, just work around it directly in ApiSandbox with document.activeElement.blur(), or do something in oojs-ui so "submitting" buttons can be configured to take focus, or maybe oojs-ui already has a workaround that I just don't know about. [15:32:12] anomie: the thing with buttons not taking focus on click predates me, and i've never been able to get a straight answer out of anyone why we do that. but you can disable it with .static.cancelButtonMouseDownEvents = false. [15:32:26] (subclass ButtonWidget or whatever, and set MyButtonWidget.static.cancelButtonMouseDownEvents = false) [15:33:14] Yeah, but that requires subclassing ButtonWidget to work around a bug, which isn't terribly developer-friendly. [15:33:46] a misfeature. but true. [15:34:04] maybe we could remove that. i honestly don't know right now. [15:38:34] * anomie decides to just work around it for now with https://gerrit.wikimedia.org/r/273243 [21:18:30] bd808: I posted at https://ur.wikipedia.org/wiki/%D8%AA%D8%A8%D8%A7%D8%AF%D9%84%DB%82_%D8%AE%DB%8C%D8%A7%D9%84_%D8%B5%D8%A7%D8%B1%D9%81:Obaid_Raza#Please_attend_to_your_bot about Obaid-bot. [21:19:07] fun rtl stuff in your sig there [21:19:50] rezabot has 4 different user-config.py files in 4 different directories :/ [21:29:41] logging in to wikitech from an incognito window logs me out from the normal window [21:29:47] is that normal? [21:30:57] tgr: I know logging *out* in one window will log you out everywhere. [21:31:17] anomie: sure, that's a feature, sort of [21:31:27] logging in shouldn't do that [21:31:37] and I'm pretty sure it doesn't in other wikis [21:31:52] Actually, since it's wikitech maybe not. That's not CA. [21:32:31] there is some weirdness with OpenStackManager which breaks your login after a few days [21:32:59] you are still logged in but don't have access to the management interface anymore [21:33:19] I was just wondering if this is the side effect of the same thing or I found a new bug [21:35:37] No idea. [21:36:49] tgr: I've heard complaints of that in the past, but never reproduced or investigated it. [21:37:19] (getting logged out when logging in from another device) [21:42:14] tgr: did all the patches for reducing memcached usage land in wmf14? I'm not seeing a drop :/ [21:42:51] ori: all but one which is still not merged ( https://gerrit.wikimedia.org/r/#/c/272776/ ) [21:43:14] * ori reviews [21:43:27] should I put that into the swat? [21:43:31] it seems low-risk [21:44:14] could do it sooner, if greg-g agrees [22:03:38] * anomie submits the patch to have load.php start warning about session use [22:10:08] Oh, duh. aude, will https://gerrit.wikimedia.org/r/#/c/272952/ make it into 1.27.0-wmf.15? [22:26:04] anomie: we will need to backport it [22:26:30] we normally deploy new wikidata code every other week, and we deployed this week [22:27:08] aude: I noticed that https://gerrit.wikimedia.org/r/#/c/273372/ will log warnings until the Wiki*data* extension is updated, even though Wiki*base* has been fixed. [22:27:55] anomie: this needs https://phabricator.wikimedia.org/T128066 fixed for us to update "Wikidata" extension [22:28:11] i have patches for that issue [22:29:37] dependency hell, ugh. [22:31:18] yeah :( [22:58:06] anomie: even with the latest patch, 18 fetches of WANCache:v:centralauth-user-... for en:Main_page (just the html req) [23:10:13] does anyone care? should I just stop mentioning it, because everyone has moved on? [23:17:43] ori: we certainly care, it's just that a lot of important stuff has been happening recently in parallel and it's easy to get distracted [23:18:45] are we pretty sure that we've backported all the .15 changes to .14? [23:18:49] I have dropped a lot of things on the floor in the past couple of weeks, so I understand. [23:19:50] speaking of too slow... ori is adding getIP() to requests that are sending log events too slow? -- https://gerrit.wikimedia.org/r/#/c/273376/ [23:20:08] the ips we see in logstash right now are pretty useless (varnishes) [23:20:43] Nothing is too slow a priori, it depends on the value and the alternatives. Have you seen https://gerrit.wikimedia.org/r/#/c/270301/ ? [23:20:57] tgr is right that getIP is not cheap [23:21:51] Yeah i've seen the hit caused by ipset in timing graphs [23:22:06] ori: unless I'm misreading something the last patch is about centralauth-session-XXX keys [23:22:15] ie. the central session, not the central users [23:22:30] https://gerrit.wikimedia.org/r/#/c/272776/ , that is [23:22:50] ah, right [23:23:17] Yeah i've seen the hit caused by ipset in timing graphs. the log processor would only fire when a log event has been selected to go to udp2log/logstash. [23:24:25] when does that happen? do we defer that to post-send? [23:24:50] we don't at this point except for the logs going to kafka [23:25:19] we probably could defer always [23:25:32] as long as we run before GC/teardown [23:25:44] yeah, which postsend does [23:26:03] should I file a task? [23:26:43] we'd just need to make a config tweak. There is logic for buffer=>true [23:27:02] that's awesome, it would be a nice win, I think [23:27:18] but... actually that won't defer processors I don't think [23:27:42] processors run after a handler has said "yes I'll log this" [23:28:05] and before the handler gets the log event [23:28:21] the buffered handler would defer the udp sends [23:28:29] but not pre-processing [23:29:12] so it's good for slow handlers but not slow processors [23:31:15] yeah. not gonna help much [23:32:11] do we have another code path that unwinds xff with less cost? [23:36:06] I'm going to pull the patch out of swat. maybe we can try it next week or some time when we aren't already concerned about perf regressions [23:43:53] bd808: just log the xff header? [23:44:52] tgr: not very fun for dashboards [23:45:23] you could do the processing in logstash [23:45:34] probably a waste of your time though [23:45:58] yeah not that excited about it [23:46:17] but getting the ip shouldn't be slow honestly [23:46:25] * bd808 was the start of that being slow [23:46:57] the trusted host list used to be single ips and I tried to make it work with CIDRs [23:47:00] bd808: thanks (for pulling it out) [23:47:24] do we do phrasing here? [23:47:36] would help? I haven't looked closely [23:47:41] (Archer reference) [23:48:27] bd808: we should start :D [23:49:00] and, /me hands bd808 a pamphlet [23:49:09] I stop myself at least twice a day from typing "that's what she said" because I'm a horrible person [23:49:41] anomie: is https://gerrit.wikimedia.org/r/#/c/273372/ blocked on wikidata? [23:49:52] I mean, is merging it to master blocked? [23:54:01] ori: it *looks* like that patch would help, yes. I'm not sure I understand bblack's comments about structure concerns [23:54:15] ori: would change the speed much [23:54:46] the main point of that patch is to avoid several hundreds of layers of function call recursion [23:55:05] which insta-kills hhvm on the ci servers [23:56:00] ...would *not* change the speed much [23:56:14] anyway, not sure if getIP hits that issue [23:56:33] the test ips are pretty perverse I think [23:56:50] to torture test the structure [23:57:06] all of the code in IP is pretty suspect too [23:57:13] so much regex [23:58:05] the IPSet call with lots of recursion is from isConfiguredProxy, I don't think that's called normally