mozilla :: #taskcluster

15 May 2017
12:02wcostajmaher|afk: ping (when you are online)
13:47catleedustin: did you say that it's normal for all tasks to have self-dependencies when inspecting them?
13:48dustinonly tasks with self-dependencies :)
13:48catleedo tasks submitted with the scheduler api get self-dependencies added?
13:49dustinthe scheduler service used `defineTask` which implicitly adds a self-dependency
13:49dustin(instead of createTask, which doesn't)
14:10jhfordaki-away: i just released 1.3.1 of the client for you
14:23g-khi, who are the security points of contact for taskcluster?
14:23dustinformally, the mozilla security contacts
14:23dustininformally, I'm happy to talk
14:25g-kthanks! are you ok having the web bug bounty reports for taskcluster assigned to you and garndt?
14:25* garndt waves hello
14:27dustinhopefully someone will find something interesting eventually
14:28dustinI think every bounty request I've ever been copied on is "OMG LOOK I got access to something intentionally public"
14:37jmaher|afkwcosta: ping
14:38wcostajmaher: I am not sure if I got you reply in bug 1363101
14:38firebot FIXED, recent osx opt BBB change turned off new talos perf-reftest job
14:40jmaherwcosta: heh, that is sort of cryptic
14:41jmaherwcosta: so it looks like there are tests that are run on esr52 and mozilla-beta, but not run on mozilla-central
14:41wcostajmaher: for example, you mention gl1 gl2 gl3 are unique for beta, but they run on m-c too, don't they?
14:42jmaherwcosta: I didn't see them on osx opt on m-c
14:42jmaherthey should be run unless there is a bug for that
14:45* wcosta checks
14:48wcostajmaher: I don't know why they don't run
14:48jmaherwcosta: I only see mochitest-webgl-e10s-X, not non-e10s
14:49wcostajmaher: in this case, it does what config file says what it must do
14:50wcostaand hg blame says I am the person that changed that hehe
14:52jmaherwcosta: heh; I often end up blaming myself
14:54wcostajmaher: I remember this patch we fixed a bunch of test scheduling, I believe there is a reason for that
14:55jmaherwcosta: in ~12 weeks we will be not running non-e10s webgl; so maybe this is a peek in the future
14:57wcostajmaher: yeah
14:58wcostajmaher: iiuc, config files in test already cares about esr52 and beta
15:23wcostadustin: ideas on how validate beta and esr52 configs?
15:23dustinwhat does "validate" mean?
15:27garndtI think by ensuring the taskgraph is what it should be before uplifting the changes for in-tree scheduling (basically uplifting all of taskcluster/* I think)
15:29jmaherwcosta: let me know how I can help with that; typically I look at a run by pulling all the treeherder jobs and looking for ones that are unique to revision A or revision B
15:30dustinthat sounds like a fast trip to a tension headache :(
15:30dustinall that squinting
15:31dustinwcosta: maybe generate a full taskgraph JSON before and after and compare?
15:31jmaherdustin: it is scripted, a short list
15:31dustinwell, maybe start with just a task list
15:31dustinand diff that
15:32dustinthen look at some specific tasks' JSON before and after to see if anything scary has changed
15:32jmaheroh, that seems reasonable
15:32jmaheryou could download that from the decision task
15:32jmaherstill code to write
15:35dustin yeah
15:35dustinyou can generate it locally too with different parameters.yml's
15:36* wcosta got lost
15:39jmaherdustin: but getting esr52 would be hard to do locally
15:39jmaheroh, with different parameters.yml- would that work really for a different branch?
15:39dustinyeah, it'd take hg a long time to check that out, I guess
15:41jmaherdustin: thanks for fixing the SETA stuff over the weekend; I cannot believe I ended up mixing True and False up
15:41dustinhehe, no problem :)
15:43garndtjmaher: you had a 50% chance of getting it right! :)
15:43jmahergarndt: lol
15:52garndtjhford: I remember I had to roll back some changes to the provisioner that reverted your changes to switch off influx use...did that get put back in now?
15:52garndtmeaning, can we kill influx now?
16:07aki-awayjhford: thank you!
16:34ochameauhi #tc!
16:35ochameauI was wondering if it is possible to define dependencies between task without involving a decision task?
16:35ochameaulike Task A should be ran before Task B
16:35camdEli: are you good with my latest changes on ?
16:39dustinochameau: via tc-github, no
16:44Elicamd: go for it ++
16:44camdEli: awesome, thanks man! :)
16:48ochameaudustin: can we craft a decision task from tc-github?
16:48dustinochameau: yes, basically taskcluster-github would start the decision task, and it would create the child tasks
16:49ochameaudustin: is there any example I could look at? or some docs if there is none
16:51* dustin tries to remember which project is using one
16:51dustinbstack: ^^ do you remember who's using a decision task?
16:52dustinthe gecko decision task is *not* a great example because it's very sophisticated and designed for gecko
16:53bstackMaybe it was balrog?
16:53bstackI'm not sure either
16:56ochameaudoesn't seem lik balrog does, they opened a bug to request what I just asked ;)
16:57dustinoh Osmose's thing? .. can't remember the name..
16:58dustinnormandy -
17:01ochameaudustin: great, thanks, I'll read that!
17:23mythmonochameau: if you have any questions about that decision task, i wrote most of it, and can help out
17:23dustinthanks :D
17:27ochameaumythmon: ok, I'll let you know.
17:48akiwcosta: are you ready for a puppet merge?
17:48wcostaaki: yep
18:37akiafter moving to claimWork, we're seeing a ton of TaskclusterRestFailures. is this normal?
18:40garndtjonasfj: ^
18:41dustinwhat are the underlying errors?
18:42jonasfjaki: timeouts?
18:42akiMay 15 11:42:31 python: signing_scriptworker 2017-05-15T18:42:31 WARNING - <class &#39;taskcluster.exceptions.TaskclusterRestFailure&#39;> Unknown Server Error
18:42akithat&#39;s what i have
18:42arrwcosta: I&#39;ve been seeing a lot of &quot;Could not retrieve catalog from remote server: Error 400 on SERVER: cannot install on Darwin at /etc/puppet/production/modules/packages/manifests/virtualenv.pp:24 on node; for all of the tc yosemite nodes
18:43arrwcosta: I&#39;m guessing they kick out that error every time they run puppet
18:45FloreHi everyone, I have an issue with mozregression involving taskcluster
18:45FloreWhen it&#39;s on bisection everything&#39;s OK, but when it switches to taskcluster, it stops working
18:45Flore4:18.64 INFO: Switching bisection method to taskcluster
18:45Flore4:18.64 INFO: Getting mozilla-central builds between 838652a84b76c273e084d0705f3f4f3be89520a8 and 8a7d0b15595f9916123848ca906f29c62d4914c9
18:47Flore4:22.49 ERROR: [SSL: SSLV3_ALERT_HANDSHAKE_FAILURE] sslv3 alert handshake failure (_ssl.c:590)
18:47garndtFlore: I&#39;m not aware of mozregression and how it works. Looking at the commit log for it, seems wlach has some some work this year...he might be able to help out
18:47FloreCan someone help me understand what the problem is and how to fix it?. I&#39;m on MacOSX
18:48dustinFlore: at a guess, you&#39;re using an old Python that doesn&#39;t recognize newer TLS certs
18:48FlorePython 2.7.10 (default, Oct 23 2015, 19:19:21)
18:49Floredustin: it&#39;s the version that comes with MacOSX
18:49FloreI tried to install python 3, but I had so many problems that I went with the preinstalled version
18:50dustinyeah, the script probably wouldn&#39;t work in 3
18:50dustinwell, that&#39;s my only guess :/
18:51FloreSo it should work with my version, except it doesn&#39;t ?
18:52dustinI don&#39;t remember exactly what version of Python finally got a working TLS
18:52dustin2015&#39;s kinda old :)
18:52akii think it varies by build settings
18:54FloreSo 2016
18:54Floreafter my build, that&#39;s the reason
18:57akihm, my problem may be polling at a 5s interval, using claimWork which holds the connection for 20s
18:57jonasfjaki: claimWork does long-polling for 20s, so sometimes it&#39;s bound to hit heroku 30s limit and result in unknown errors... but I&#39;m also a bit concerned with the error rate
18:57jonasfjaki: for sure you don&#39;t want parallel polls :)
18:59akino, that&#39;s probably not it, because we wait for it
18:59jonasfjwhile not self.stopping:
18:59jonasfj claims = self.queue.claimWork(..., {tasks: 1})
18:59jonasfj for claim in claims.tasks:
18:59jonasfj self.processTask(claim)
18:59jonasfj sleep(5s)
18:59jonasfjaki: something ike that ^
19:00jonasfjyeah, then that&#39;s close to what you have :)
19:00akinow i&#39;m back to not knowing what&#39;s going on
19:01jonasfjaki: I think it&#39;s queue side issues..
19:02jonasfjaki: if you can print more information from the exception like the body that would be interesting...
19:02akishould i catch those?
19:02jonasfjif polling fails... try again... :)
19:02akii think is eating the error, if any:
19:02jonasfjI&#39;m not sure we have a quick solution for this...
19:03jonasfjyeah, the python client isn&#39;t good a propogating error information..
19:03jonasfjthe node one attaches pretty much everthing to the error object
19:04akiwe might want to get the data into the message
19:04dustinopen an issue?
19:04dustinI think this isn&#39;t the first time we&#39;ve bumped into this :(
19:05jonasfjaki: in node I always log errors with all the properties... that helps... messages otherwise risk becoming way too bug..
19:08catleepy3 has a nice &#39;raise FooError from e&#39; feature for exception chaining
19:10akiyeah, supports py2 and it looks like we&#39;re trying to keep the async and sync logic the same
19:11akirequests and aiohttp behave slightly differently so we&#39;re not using their raise_for_status
19:25jonasfjaki: wait that caused TaskclusterRestFailure ?
19:25akii think so
19:25akiupdating the client and if this was it, the errors will go away
19:58wcostaarr: sorry, was afk
20:02dustinall of these windows tests I&#39;m debugging turn green when you look at them :(
20:32camdEli: is there an easy way to start yarn/neutrino using a different port? it&#39;s conflicting with treeherder on my system
20:38dustinI think the port is in your app&#39;s package.json?
20:41garndtKWierso: thanks for summarizing that!
20:43Elicamd: PORT=4500 neutrino start
20:43Elior whatever
20:43Eliyou can make it permanent with config overrides
20:44camdEli: ahhh, gotcha. thanks man! :)
20:45Elinp :)
20:55FloreHello, for those interested, I finally managed to make mozregression work by installing python2.7.13 with homebrew :)
21:00Floreyep, so the problem was indeed a too old version of python
21:00dustinlucky guess :)
21:00FloreWell it worked, so thank you :)
21:18hwineanyone remember where the announcement about moving tcbot from an OAuth app to github integration was? Google can&#39;t find it :(
21:27bstackhmm, I can&#39;t find it either
21:27* hwine wonders if it was email only
21:29hwineAh, yeah - Feb 3 email from Irene
21:32dustinthere was a blog post too
21:33hwineahh -- I searched for &#39;taskcluster bot&#39;, and that term doesn&#39;t exist anymore ;)
23:53jryanshmm, what are my options for getting gdb onto a one click loaner? there appears to be no package manager available.
16 May 2017
No messages
Last message: 13 days and 14 hours ago