mozilla :: #taskcluster

7 Sep 2017
06:09pmooredustin: good question, I'll need to check
06:09pmooregood point about retriggers
07:17igoldanI see the load on talos-linux64-ix is still increasing
07:17igoldancurrently, it's at ~1.700 jobs
07:18igoldanshould this be looked over?
08:29aobreja|builddutyigoldan: I'm checking now ,the problem seems to be with the machines that are not taking any new jobs,will come with updates
08:45jmaher|afkigoldan: I see chatter in #releng about the linux machines
11:29igoldanlooks like there is a bug on the Backfill functionality:
11:30igoldanis this known, or should I file this on Bugzilla?
11:48jmaherigoldan: I believe that is before the fix landed
11:49jmaherigoldan: so the old trick of editing the task
11:53jmaherodd, I pushed to try to test moving some jobs from win8-win10, the confusing thing is I am getting jobs run in double
11:54jmaherin the target-tasks.json seen in the decision task, I see only one instance of the job; maybe this is something inside of buildbot?
11:56jmaherigoldan: were you able to figure out how to add jobs to backfill for that ?
12:00igoldanjmaher: yes, I simply used "Add new jobs"
12:01jmaherigoldan: ok, sorry for all this confusion, I would like to think we are almost done with regressions that have occurred during the "dark ages" of backfilling
12:46dustinI would very much like to think so too :)
13:04dustinpmoore: so, superseding..
13:05dustinI am trying to document what we're building, rather than reopen the question of the philosophy of superseding :)
13:11jonasfjpmoore: if one day we have the data store to implement superseding in the queue, I think you can win that argument :)
13:11* dustin keeps it on the issue
13:12dustinsorry to step on you jonasfj
13:13jonasfjplease tune in... I'm opening mail from last week..
13:13dustinthanks for not just deleting all of it
13:14garndtHis delete filter only works on mail from me
13:14jonasfjI never delete anything, I archive it for reading in 100 years when Google uploads my brain to the cloud
13:14dustinman, if I could mentally mapreduce
13:14dustinthat'd be AWESOME
13:15jonasfjI would duplicate myself to each of me only have process one email...
13:18jonasfjdustin: randomly, I agree a resilient/repair mode for azure-entities would be nice..
13:18dustinthat Provisioner-in-TaskDependencies thing was kind of a nightmare :(
13:22jonasfjyeah... it's a mess
13:22jonasfjI'm still tempted to suggest we invest in postgres instead...
13:23dustinI agree, although I don't think we have the resources to do that righ tnow
13:24dustinbtw two things I think we should do before diving into postgres: (a) send someone to a postgres training, (b) establish some ground-rules similar to the constraints imposed by azure
13:24dustin(c) write a library that enforces at least some of those rules
13:26dustinBuildbot's strategy (after trying a few other strategies) was to write wrappers for every "entity" that encapsulate the complex queries
13:26dustinso the joins, etc. are all in one place and easily examined for efficiency
13:26dustin(it also made us backend-independent -- BB runs great across mysql, postgres, and sqlite)
13:27jonasfjI agree, we need something like azure-entities
13:27dustinyeah, with an escape hatch for implementing more complex queries only as required
13:27dustinno ORM
13:28garndtOk ok then Taskcluster-activerecord
13:28jonasfj ^ who doesn't want an optimistic storage solution :)
13:29jonasfjthat's a good logo
13:29jonasfj / mascot
13:29jonasfjsomeday, when we have too much time..
13:30dustinI think it's a good mascot for me :)
13:40pmooredustin: i have a meetings at 7am, 8am, 11am Pacific today - but if you like we can meet on vidyo between the 8am one and the 11am one to discuss superseding stuff
13:42pmoorejust send me a meeting request
13:46pmooreaki-away: if you randomly happen to be around in 15 mins, we'll be having a discussion in wcosta's meeting room about reducing pain in AMI rollout (i know it will be like 7am for you)
14:01wcostapmoore: are you coming?
14:03dustinpmoore: want to meet now?
14:04dustinmixed my timezones up
14:29firebotBug 1395699 NEW, [meta] Automate workers deployment
14:46gerard-majaxpmoore, hello
14:46gerard-majaxpmoore, do you know who worked on yaml tempates for gecko ?
14:47jonasfjgerard-majax -> dustin
14:48dustinjlal long ago
14:48dustinuhoh everyone look busy
14:48selenamarie:) i know. when i show up there is trouble.
14:48gerard-majaxjonasfj, thanks
14:48gerard-majaxdustin, I have a quick question indeed
14:48selenamariebut actually, i just have a practical question: when someone asks about a build, where does the bug go?
14:49selenamariei see in Core::General
14:49jonasfjdepends on what they ask..?
14:49gerard-majaxdustin, how hard would it be for me to re-use the tooling with templace inheritance that lives in gecko for a tc-gh repo ?
14:49selenamarieasking about a mysterious ASAN build
14:49selenamarieTask Configuration?
14:50dustingerard-majax: there's no template inheritance in gecko
14:50jonasfjasking: why it's build? how to build it? why it doesn't work?
14:50dustinjonasfj: they're asking why a bulid stopped being built
14:50selenamariejonasfj: why isn't the build running anymore
14:50gerard-majaxdustin, there was in the past, was that removed?
14:50dustingerard-majax: yes
14:50dustinyou're welcome to use it
14:50dustinit was taskcluster/taskgraph/util/ or the like
14:51dustinit *might* still be in there - ISTR there was one thing still using its merge functionality, and I don't recall if I fixed that
14:51jonasfjyeah, task configuration is a good place to ask... and hg blame to who did it...
14:51dustinselenamarie: so I tihnk we could figure that out from hg history
14:51jonasfjgerard-majax: are you looking to add a decision task to a gh repo?
14:51gerard-majaxjonasfj, yes, exactly
14:52gerard-majaxjonasfj, I already have one, technically
14:52selenamariedustin: cool. i'm just clearing out Core::General. enjoy!
14:52dustingerard-majax: I'd suggest using json-e instead if possible
14:52gerard-majaxjonasfj, but I did that quickly, and I am planning to rethink it for a cleaner and easier maintainance
14:52jonasfjoh, cool
14:52jonasfjyeah, I dream of having some way to do it elegantly... we don't have that yet..
14:52gerard-majaxI remember that someone mentionned json-e already but it was not yet ready I think
14:52dustinthe templates stuff wasn't bad, it just wasn't powerful enough for the generation process we needed
14:53dustinjson-e is ready now
14:53gerard-majaxjonasfj, right now it's .yml file with alike-templates handled by sed :)
14:53jonasfjthen json-e is much better...
14:53dustinbetter than sed for sure :)
14:54gerard-majaxdustin, ?
14:54dustinyou can implement "inheritance" in it with $mergeDeep
14:54gerard-majaxwoo, that seems non trivial
14:55jonasfjor even better...
14:55dustindepends how you want to set it up I think
14:55gerard-majaxwell that's the low-level foundations
14:56gerard-majaxI guess I will have to play with that a bit
14:56gerard-majaxand that is the code ?
14:58gerard-majaxdustin, jonasfj, thanks, I'll look into json-e use then
14:59jonasfjI dream of something like:
14:59dustinmeeting time
14:59* dustin wakes up napping cat :( :(
14:59arrdustin: do you still hav elimbs left?
15:00dustinyes :)
15:01gerard-majaxjonasfj, I'm not sure I completely follow you
15:01arrdustin: not so bad, then :D
15:02jonasfjthe idea is to have json-e with a function render
15:02jonasfjwhere "render" is an injected function that loads a template and renders it with given config
15:03bstackwho is this jonasfj person in this channel?
15:03bstack(but also welcome back)
15:03dustinhe looks so relaxed
16:54dustingps: hm, MozbuildObject.mozbuild_reader.. how do I get a MozbuildObject?
16:54dustinis there some global where the current mach command is stored? (*hopeful*)
17:56gpsdustin: on the @Command function in, use `self.mozbuild_reader`
17:57gpsdustin: you can also call MozbuildObject.from_environment() and it attempts to do the right thing
18:00dustinI'll need to do the latter -- this is pretty deeply buried
18:26gpsi'd like to use --privileged docker contains so i can use chroot() within a task...
18:26gpsi could probably hack something together by abusing dind-service. but i suspect someone here will advise me against that.
18:27dustinsounds like a job for jonas
18:27dustinwe have a ptrace-related feature that's only enabled on certain tasks
18:27dustincould do sometihng similar here
18:28gpsthat would be epic
18:36dustinah, so it'd be useful to use the debian tooling
18:36dustinugh, it was months of therapy to forget that stuff
18:36dustindebian packaging is the WORST
18:37dustinanyway, if there's an AppArmor profile we can run that would allow what you need without the full --privileged, that would be best.
18:37gps isn't so bad
18:38gpsmost of that YAML can be derived from Packages files
18:38gpsand i would think that 90+% of packages we build would be direct from upstream, so we wouldn't need to worry about Debian packaging muckery
18:39dustinthat's good
18:39gpsactually, it appears we can use `fakechroot` with debootstrap. so i may not actually need chroot()!
18:40gpsheh. mercurial has a debian scaffold in its repo. that's what i used
18:40gpsi just renamed the package "-mozilla1" and added robustcheckout and our custom hgrc mods
18:42catleedebian's packaging is bad, but not as bad as rpm!
18:42dustinlet's just agree they're both bad :)
18:42catleelet's make a better one
18:44dustinhaha, I know when I'm beaten
19:16arrdustin: garbas is out till october
19:16dustinyeah, sorry, I meant to say so in the email
19:16dustinarmenzg_brb: ^^
19:17arrarmen started his email saying that :D
19:17dustinoh, then I didn't have to mention it :)
19:21armenzg_brbno worries
19:21armenzg_brbdustin: I see the header being set
19:21armenzgtbh the shipit-uplift app was hijacked by marco (according to garbas)
19:21dustinwhta's the API URL?
19:21dustinthat seems bad
19:22armenzgthe API is not yet available
19:22armenzgbut a similar one is here
19:22armenzgwhich means it will have the same CORS restriction
19:22armenzgmy React UI is currently hosted here
19:23dustinso it seems like marco needs to host this in the right place :)
19:23armenzgshould I be looking at changing Access-Control-Allow-Origin to *?
19:23dustinand set his CORS headers correctly
19:23dustinif * is really appropriate
19:23armenzgdustin: any suggestions?
19:24armenzgthe convincing arguments?
19:24dustinthat it's not ship-it
19:24dustinwhether it should be part of releng services, I don't know
19:24dustinbut at least it should be its own releng-services app
19:24armenzgI thought this was the wild west :P
19:25armenzgI find this releng-services rather heavy handed
19:25dustinwell, it's not my area so i can't say :)
19:25armenzgI like Heroku
19:25dustinso if marco and you are both writing different parts of the same app (coverage)
19:25dustinthen it makes sense to host them on the same domain
19:25dustinand maybe not be a releng service
19:25dustinand as for CORS
19:25dustinif the data is intended only for use by pages loaded from the same app, then the existing CORS is correct
19:26dustinbut if the data is meant to be consumed by any web page, then * is correct
19:26dustinhas this had an RRA btw?
19:27armenzgdustin: if there's a backend app on Heroku and a frontend one on Heroku as well; does it also require CORS?
19:27dustinwell, CORS is always enabled in a browser :)
19:27armenzgdustin: I doubt it
19:27dustinit's just different configuration
19:27dustinthis should probably have an RRA :)
19:27armenzg(wrt RRA)
19:28dustinonce it's at the proof-of-concept stage anyway, but before people start using it
19:29armenzglet me poke at things
19:29armenzgI think that API exists because this other app
19:29armenzg(shipit instead shipit uplift)
19:30dustinso it is sort of shipit related
19:31armenzgI think it also got shoehorned
19:31armenzgwell, probably relman uses it
19:32dustinseparation of concerns is good :)
19:34tedTIL how to query task runtimes in ActiveData
19:34tedpretty useful!
19:34dhousedustin: I'm going to check-in Jake's patch as-is for the intermittent osx virtualenv failure finding python.
19:34firebotBug 1396266 REOPENED, Intermittent OSX "The executable '/tools/buildbot/bin/python' is not found; not creating virtualenv!
19:35dustinsounds good
19:35dustinI thought he had landed it tbh
19:35dustinoh, right, I r-'d it
19:35dustinyeah, that's fine
19:35* dustin paging memories back in slowly
19:35tedi was able to answer my question "did the /home -> /build change fix the mac debug gtest slowness?"
19:35dhouseyou r+'d with a suggestion ;)
19:35dustinted: that's great!
19:56ahalwe should build a nicer frontend around active data
19:56ahalI think it would get used a lot more
20:19ted i find the query syntax a little offputting, to be sure
21:20Elidid someone say frontend?
21:20Eli*a wild Eli appears*
21:21arrdoes a wild eli have twigs in its beard? :D
21:21arror not that kind of wild? :D
21:22Elii barely have a beard, so sadly no
21:22* arr steals some of garndt's for you
21:23arrEli: you just made me lol
21:24garndtfeel free to take some beard, I have plenty of it
21:24Eligarndt: GET OUT OF HERE
21:24garndthush you
21:24* Eli grumbles *PTO hypocrite*.
21:25dustinwe need a pto kickbot
21:26garndt /kick garndt get out you
21:26* garndt waves
21:26garndtI was never here
21:26arrgarbas: set a good exmaple!
21:26arror example
21:27dustinor garndt
21:27arromg *is* setting a good example
23:00dustinI love that a company I can't choose not to do business with did this.
23:03dustinthis was not the channel I meant to bring that up in
23:03dustinbut there you are :)
8 Sep 2017
No messages
Last message: 14 days and 2 hours ago