mozilla :: #taskcluster

8 Sep 2017
00:02gpsi want to know how got a checkouts cache that had run-task metadata but didn't have a VCS cache
00:03gpsi guess that means somewhere we're adding the cache to a task that doesn't do a checkout?
01:47dustinor the task failed midway through?
02:47glandiumwhat's going on?
02:58firebotBug 1336178 NEW, Intermittent [taskcluster:error] Task was aborted because states could not be created successfully.
02:59philoror not, in truth I never read the rest of the message to know whether it's the same :)
03:04garndtglandium: if you rebase it should clear it up. There was a bad commit that broke somethings but should be fixed now
03:06garndtI'm sorry :( I know it's a pain. It wasn't broke on purpose but hopefully fixed for good now
12:36jonasfjdustin: don't user need mozilla-group:scm_level_{1,2,3} in their certificate?
12:37dustinwe have mozilla-group:active_scm_level_{1,2,3} now
12:37dustinwhere are regular LDAP groups
13:00dustinI tried logging in after landing that and it worked OK
13:00dustinI'll delete the old groups Monday (although I think deleting them now wouldn't hurt)
13:00dustin*old roles
15:56garndtcatlee: do we have anything in AWS that's still running in buildbot?
15:56garndtthat still needs to be migrated over
15:57dustinI think so, per convo with arr this mornign
15:57dustinI tihnk there are ESR buliders/testers still in AWS
15:57dustin(..that are using proxxy)
15:57garndtI thought we were leaving ESR alone
15:57dustinright, it doesn't need to be ported
15:57arrI don't know what TB is doing, because I'm prettu sure that's still on buildbot, too
15:58arrcatlee: would have better state on what's still running in buildbot than me, though
15:58garndtand we probably have more windows 10 stuff to turn on, right?
15:58garndttrying to get an idea if there are more things that will be enabled in TC by year end
15:59arrgarndt: yes, definitely more w10
15:59garndtdo we have an estimate of how much more?
15:59garndttrying to get an idea of how much our AWS spend will change over the next few months or are we starting to stabilize for the most part
16:00arrgarndt: that spreadsheet randi is using to track tests will tell you which tests. I don't know how many resources each of those tests use (302 jmaher)
16:01catleearr, garndt: yes, we're still running some win7 GFX tests on buildbot
16:02catleeignoring ESR52 and TB
16:04garndtcatlee: thanks!
16:05catleenow I'm not sure why any more...
16:07catleeWindows 7 VM-GFX 32-bit autoland debug test reftest-gpu-e10s
16:08catleethat's the only one I think
16:09catleejmaher: does that test need to run on BB?
16:10catleeoh right, it was breaking the TC instances when we tried there
16:13firebotBug 1356240 NEW, Win7 reftest-gpu-e10s not being scheduled on Taskcluster
16:51jmahercatlee: I don't know about the reftest-gpu job, I believe it runs on hardware; that is a job that was turned on in recent times and I haven't examined it much
16:52catleeit runs on VMs right now
17:41gpsdustin: see what you made me do! :)
18:13catleeyeah, reftest-gpu-debug still kills TC Win7 instances
18:40dustingps: that's .. crazy/cool
19:01catleeare you trying to pin/rebuild all the deps?
19:01dustinI love that gps thinks no small thoughts
19:02dustinwell, he probably occasionally thinks "I'm gonna go get a snack" but 99% of the time big thoughts
19:02catleethe big thoughts happen while he's having a snack
19:08catleejmaher|afk, arr: filed for the windows 7 debug reftest failure
19:08firebotBug 1398284 NEW, relops@infra-ops.bugs Debug reftest e10s tests break Windows 7 GPU instances in Taskcluster
19:09catleethat's the only test we're running in BB now for win7
19:10catleegps: you should show that to some debian folks :)
19:16garndtcatlee: do we have any windows 7 on hardware that's being migrated or is windows 7 only AWS?
19:17jmaher|afkgarndt: we still have mochitest-chrome, mochtiest-clipboard
19:17garndtjmaher: ok, so those will be migrated to new hardware when we have it?
19:17garndtthanks sir
19:17catleegarndt: yeah, talos is still on hw
19:17catleeand those two
19:17garndtok, talos too, great
19:19garndtdo we know if any of these tests have been tested on the new hardware yet? or will it be a bunch of unknowns
19:19garndtjmaher: is talos the only hw jobs we have for linux?
19:21jmahergarndt: as far as I know
19:21arrno graphics for linux?
19:22jmaherarr: they all run on vm iirc
19:23arrI'm all for putting more stuff on the vms :}
19:23jmaherwe use llvm-pipe and related tools on the ubuntu vm
19:23dustinmeaning EC2 in this case?
19:24jmaherI guess so, is that amazon aws and what we run on taskcluster docker stuff?
19:25dustinthe tsuff that used to be builders with "VM" in the name
19:25dustinI just want to be sure we're not talking about vmware, or Xenserver on the moonshot hardware ,which are both also VMs :)
19:25jmaherso many choices
19:26dustinyeah, it's crazy
19:29garndtarr: do you know if any of the windows 10 tests were tested on the new hardware with generic-worker to see if they are green or not, or when we get the new hardware we will need to spend time greening them up?
19:30arrgarbas: markco is doing that now
19:30arrgarndt: ^^
19:31arrwell, at the moment, he's focusing on w7 since there are some unexpected issues there
19:31arr(the OS sees the graphics card, the tests do not)
19:31garndtwe don't need no stinkin' graphics cards!
19:31arrbut we ran some tests on them back before we bought the hardware
19:32arrI doubt it's all going to be green since things change :}
19:32arras soon as mdc1 (physical) comes online, we'll have a lot more infra to green tests on, though
19:33arrshould be 125/125/100 (w10/w7/ubuntu)
19:33arrer 125/125/50
19:34arrno, 100
19:34arrI was right
19:34arrmath is HARD
19:34arr300 nodes total :}
19:35arrI'm sure we'll reserve a couple of those so that we can debug any deployment problems that crop up, but there should be lots left over for test greening (as much as we ever have "lots' of something that's not dynamically scalable)
21:27gpsi found another hack to make I/O faster (bug 1398355)
21:27firebot NEW, Use eatmydata when I/O correctness isn't required
21:27gpsonly a 3x speedup for Docker image building
21:31garndtthat's pretty neat
21:31gpsyet another reason to have a unified docker image builder - so things like this can get injected early and consistently
21:32gpsdustin: ^
21:32dustinlol, that seems like a "great" idea :)
21:33gpsfor a lot of our tasks, we don't care about I/O persistence or durability
21:34gpstmpfs would be even faster/better
21:37gpscan we get a worker feature to mount tmpfs volumes?
21:37gpsthat would be a P1 performance feature
9 Sep 2017
No messages
Last message: 11 days and 4 hours ago