mozilla :: #testpilot

14 Mar 2017
00:40fzzzyI was so much happier when we moved pageshot into a normal browser window and out of a panel :)
01:36_6a68our DAU is _so close_ to 100,000 users! https://irccloud.mozilla.com/file/pGaIAoAz/Screen%20Shot%202017-03-13%20at%206.35.09%20PM.png
02:00fzzzynice!
14:09ckpricewe pushed stage last night, looks like it broke a couple things #2317, 2318
14:09ckpriceboth look to be l10n related
14:09ckpricedev seems unaffected, and the diff between stage-static and master don't show anything too crazy
14:41ckpricefzzzy: any clue on 2317/2318?
14:41ckpricejgruen: clouserw fyi ^
14:41jgruenyeah, i am digging into it
14:41jgruennot sure what even landed in the last two weeks that would have caused it
14:41jgruenbut we obviously can;t push to prod like that
14:48flodgiven that everything seems to work normally on dev, can't it be just a build/deployment issue on stage? https://testpilot.dev.mozaws.net/
14:49floddev is localized correctly for me, while stage is in English with an empty green button like 2317
14:49jgruenflod, yeah, that's kind of what i'm thinking
14:49jgruenonly PRs on dev not on stage are pontoon
14:49jgruenckprice: want to do another stage deployment and see what happens?
14:49jgruenmaybe some weird caching in stage or something?
15:02ckpricejgruen: comin right up
15:05cloudops-ansibletestpilot static stage #92 building
15:05ckpricejgruen: flod building
15:05ckpriceCosminMCG: ^
15:07CosminMCGckprice: ok, thanks
15:21cloudops-ansibletestpilot static stage #92 deploy succeeded
15:24flodis '2017-13-03' currently deployed on stage, or something else?
15:25clouserwstage is at HEAD right now
15:25clouserwhttps://github.com/mozilla/testpilot/tree/stage-static
15:25fzzzyhttps://irccloud.mozilla.com/pastebin/YFJz42ct/
15:26CosminMCGckprice, clouserw: it seems that the issues are still reproducible
15:27flodso, what changes between stage and dev?
15:28fzzzylooks like experiments.afb96a82.ftl is coming back as an html document instead of an ftl
15:29flodoh wait, why does it add stuff to the file name?
15:32flod(never mind, it adds that to dev too apparently)
15:34ckpricestage and dev are currently in sync
15:34fzzzyyeah, it is coming back as an html document because it is the 404 page
15:37fzzzycurl --head https://testpilot.stage.mozaws.net/static/locales/en-US/experiments.afb96a82.ftl shows: X-Cache: Error from cloudfront
15:38flodthis clearly introduced that, but locally it works fine, mess happens when deployed :-\ https://github.com/mozilla/testpilot/pull/2287
15:38fzzzyflod: right. ckprice, clouserw: is something missing from the deploy script?
15:39ckpricedid that patch update anything in deployment?
15:39clouserwI touched it recently to add a /latest/ URL
15:40clouserwrelud: is there any output from the deploy script on stage?
15:41fzzzy /static/locales/en-US/app.04647fcf.ftl and /static/locales/en-US/experiments.afb96a82.ftl are 404
15:41fzzzyckprice: that patch added links to those files
15:45flodis there a way to exclude the en-US .ftl files from caching? as far as I can tell we're not caching other locales
15:46flod(and I realize caching is not the right word here)
15:47clouserwftl files are cached for 10min
15:47clouserw(we can change that to anything we want)
15:48flodI mean avoid adding the afb96a82 part
15:48fzzzy /static/locales/en-US/app.ftl and /static/locales/en-US/experiments.ftl are coming back with the correct files on stage
15:48clouserwwithout the hash in the filename we could get weird situations during pushes
15:49flodisn't that already happening for other locales?
15:52flodsorry, I realize I'm not helping given I know zero about the deployment process, so I'll shut up. Unfortunately backing out those two lines will create other issues (IDs showing up instead of translation)
15:53fzzzymaybe we just have to wait for cloudflare to have those files instead of 404 pages
15:54flodwhat's generating the content of static?
15:54flode.g. /static/styles/main.219c2604.css?
15:55clouserwfzzzy: you think a request is hitting and caching the 404 before the file exists?
15:55fzzzyclouserw: seems plausible
15:55clouserwI was thinking the file doesn't exist ever
15:56fzzzyyeah that might be it too
15:56clouserwis it generating for you locally?
16:00lorchard|awayHmm, doesn't seem like this should be causing an issue, but we did tweak regexes for excluding some images from getting content hashes in their filenames https://github.com/mozilla/testpilot/commit/5f8dbe418bcfe36621d5551a3ee1d66a5096652f
16:01fzzzyno, locally I only see the file without the hash
16:01lorchardLocal dev isn't doing that unless you run `npm run static`
16:02lorchardOr `gulp dist-build` - either way, local dev doesn't do *all* the things we do for deployment
16:03fzzzyafter running npm run static I still only see the files without the hash
16:04lorchardDo the files with hashes come back if you revert 5f8dbe418bcfe36621d5551a3ee1d66a5096652f ?
16:05lorchardPossibly also 36ec1a34b7c0b92a261f3e804120c7cf6ee54531 ?
16:05flodlorchard: that only tries to load en-US files, I don't think en-US files with hashes are supposed to go back at all
16:06lorchardthose were places where we mucked with `dontRenameFile` for rev-all in social images
16:06lorchardThough also, I don't think we're hashing FTL files
16:07flodso, IMO the point is: is there a way to tell the system to ignore those two lines when adding hashes?
16:08lorchardYes, we should already be doing that: https://github.com/mozilla/testpilot/blob/master/frontend/tasks/dist.js#L14
16:08lorchardAnd they won't work hashed in the page template anyway https://github.com/lmorchard/testpilot/blob/master/frontend/src/templates/index.mustache#L12
16:09lorchardBecause of the URL template format in <link rel=&quot;localization&quot;> we can&#39;t hash the filenames
16:10lorchardBecause instead of {locale}/app.ftl, it would need to be something like {locale}/app.{hash}.ftl - and {hash} is a thing that varies from file to file that we don&#39;t know on the client
16:10lorchardWe&#39;d talked about hashing the *directory* name instead for locales - e.g. with a git commit hash or something
16:11flodso, but what is adding hashes to en-US but not the second two lines, if you look at source in https://testpilot.dev.mozaws.net/ ?
16:13lorchardOh, locale URLs are missing from `dontUpdateReference` https://github.com/mozilla/testpilot/blob/master/frontend/tasks/dist.js#L18
16:13lorchardbecause we hadn&#39;t used a direct reference to a locale URL up to this point
16:13lorchardSo, something like this needs to be added to that lie /static\/locales\/*/,
16:13lorchards/lie/list/
16:15lorchardrev-all hashes filenames and then replaces URL paths with hashed names - `dontRenameFile` controls the former and `dontUpdateReference` the latter
16:15* flod tries locally
16:16lorchardI missed the part about adding the en-US fallback in the channel scrollback
16:16flodweird, it doesn&#39;t seem to work for me
16:17flodoh wait, it does in dist/home/index.html, not in dist/index.html
16:20lorchardWorked for me locally
16:24flodhttps://github.com/mozilla/testpilot/pull/2320
16:24ckpriceready to repush to stage when that gets merged
16:24lorchardHeh https://github.com/mozilla/testpilot/pull/2321
16:24lorchardI&#39;ll close mine
16:25flod(sorry about that)
16:26lorchardI just merged it
16:27cloudops-ansibletestpilot static stage #93 building
16:31flodok, hashing has disappeared on dev in the meantime
16:34lorchardThe asset path hashing stuff is weird, because we&#39;ve had to have exceptions for certain well-known or inflexible URLs
16:34lorchardThe add-on moved, so maybe we don&#39;t need those particular exceptions anymore
16:37pauloiegasSVckprice, clouserw: Guys, please keep in mind that we won&#39;t be able to test Production today if you are planning to deploy it also. Maybe we should postpone the prod deploy until tomorrow so we can properly test the stage fix.
16:37flodstage looks localized correctly now?
16:37flodeven if the bot didn&#39;t say anything yet
16:39CosminMCGit looks like, but I am not sure if the deploy is finished, the localization looks fixed on stage now
16:41fzzzyi think it&#39;ll be ok. it was obviously those two files and nothing else is likely to be broken
16:43cloudops-ansibletestpilot static stage #93 deploy succeeded
16:51lorchardOh hey, I think, I figured out why I was getting a blank window for the Browser Toolbox: The window asking permission to connect was moved off screen somehow :/
16:56ckpriceCosminMCG: pauloiegasSV no problem, we will resched our prod push for tomorrow
16:56ckpricethank you for staying late
16:56pauloiegasSVckprice: no problem, glad we could help
17:16reludclouserw: stage log from 8 am 1/2 https://irccloud.mozilla.com/pastebin/ZN67y8Mm/
17:17reludclouserw: stage log 2/2 https://irccloud.mozilla.com/pastebin/um6U0Hnd/
17:18reludthe job succeeded, and so did the subsequent one. latest is in there. the log was too long for a single pastebin.
17:21reludpretty sure those are irrelevant now, but just in case
17:23clouserwthanks
17:32pdehaanhttps://public.etherpad-mozilla.org/p/snoozetabs-coodination
18:07clouserwjgruen: did you check the stats for our A/B whiteborder test?
18:07jgruennot yet
18:07jgruenin page shot meeting
19:12fzzzydcoates: can i get an r+? the only thing that changed is i put the loader bar back in and rebased and squashed https://github.com/mozilla/testpilot/pull/2306
19:12dcoateslooking now
19:15dcoatesfzzzy r+ed
19:15fzzzyty!
20:34lorchardOh huh, I think I might be making headway starting from scratch with this Toolbar / Menu styling thing in a WebExtension in case anyone&#39;s following along https://github.com/lmorchard/webextension-toolbar-vs-menu
20:38lorchardNow I have to shovel some snow while I still have daylight wheee
23:57dcoateswebextensions api is kind of infuriating
15 Mar 2017
No messages
   
Last message: 192 days and 8 hours ago