mozilla :: #rust-infra

19 Apr 2017
00:02frewsxcv has been running for three hours
00:29aidanhsfrewsxcv: it heard you :)
00:30aidanhsahh, the sleeps
01:00simulacrumWell, I got perf.rlo up on heroku (for free):
01:00simulacrumIt won't auto update, but it's a start
01:00simulacrumObviously temporary
01:01simulacrumnmatsakis: IIRC, you were interested in potentially looking at some of the data, so ^
01:04tedIcefoz: love it :)
13:31aidanhsI note that there have been attempts to set the reviewer via r? which have failed previously - just want to double check that using the github UI is the preferred way to go about setting assignee
13:46aidanhswe may be about to have a problem with PRs
13:47aidanhsseems like a url we use in the fuschia docker container has died
13:48frewsxcvWhat's the url
13:48frewsxcvDon't want to load Travis on my phone
13:49arielbycan we stop relying on networking on builders?
13:49aidanhsif I'm right, we're on borrowed time while the fuschia image can be reused from the cache
13:49frewsxcvCan you open a new issue and cc fuchsia people in it?
13:50frewsxcvI can find the people
13:51frewsxcvtedsta and raphlinus
13:54frewsxcvServo uses 'treeclosed' for bors which prevents PRs from landing. Wonder if we should start using that
13:56steveklabnikimho, closing the tree is an antipattern
13:56steveklabnikmy opinion doesn't count for much though ;)
13:57frewsxcvWhat's the point of trying to land PRs if they're all going to fail? We'll just have to go back and retry all the failed ones
13:57frewsxcvCurious to hear why you think it's an antipattern
13:57aidanhswait...maybe someone has already merged a fix for this since the rollup is going fine
13:58steveklabnikit treats the symptom not the cause
13:58steveklabnikthat is, sure, in some emergency situations, it makes sense
13:58steveklabnikbut, if it happens anywhere near regularly, it means something else is very wrong
14:01frewsxcvSure, but I don't see how tree closure makes the situation works
14:01steveklabniki'm not saying it wouldn't in this situation
14:02steveklabnikbut, if you build tools that are easy to use, people use them
14:02* steveklabnik shrugs
14:02steveklabniki'm not on infra, so yinz should decide what's right
14:03aidanhsit's a sort-of false alarm
14:03aidanhsany pr branched off longer than 12 days ago will fail
14:03aidanhsso rollups will still go through
14:03frewsxcvMerge commits should take care of that though
14:06TimNNI've found that commit in the github mirror:
14:07TimNNTo me that looks like a rebase + force push
14:08aidanhsit's a PR to beta
14:10aidanhsso I guess the url needs putting back, or the fuchsia fix on master needs backporting to beta as well
15:09acrichtoaidanhs: frewsxcv: if fuchsia is down we should coment out that builder
15:09acrichtoand cc the fuchsia guys
15:09acrichtoTimNN: ^
15:10acrichtolooking to see if it's happened already
15:10simulacrumI don't think so
15:10acrichtook I'll send a PR
15:10simulacrumacrichto: Also, you may be interested in this:
15:10simulacrum(I put it up for free for the time being)
15:11simulacrumMight even work to move the permanent solution over to heroku if we wanted to, not sure what is cheaper / easier for mozilla
15:11acrichtohey I find heroku always easier
15:12simulacrumIt took me maybe 15 minutes to get it up, and the benefit could also be that permissions are easier to give up
15:12simulacrumerr, give out
15:16acrichtoaidanhs: r?
15:18aidanhsacrichto: looks good (but pretty sure I can't r+_
15:19acrichtoaidanhs: thanks for looking!
15:19acrichtoaidanhs: thanks for diagnosing that!
15:21arielbythe most obvious thing is that it seems that LLVM times have sped up by 13% in commit 7c7753d370
15:21acrichtoaidanhs: whoa... does this look familiar? --
15:22acrichtoBUG: submodule considered for cloning,doesn't need cloning any more?
15:23aidanhsback to the drawing board with the git cache again then
15:23aidanhswhat a pain
15:24simulacrumThis seems suspicious: [00:06:22] fatal: Not a git repository: /Users/travis/rustsrc/src/src/llvm/../../.git/modules/src/llvm
15:24simulacrum[00:06:35] fatal: Not a git repository: ../../.git/modules/src/llvm
15:24acrichtoaidanhs: :(
15:24acrichtoin the long run it may just not be worth it
15:24aidanhsthe cache extraction terminated early and the integrity checks didn't pick it up
15:25aidanhsI can fix the integrity checks, but if extracting the cache is taking too long then it needs a different approach
15:26larsbergsteveklabnik: the biggest existing scenario for `treeclosed` is when you're doing a rust compiler or big upgrade that is easily bitrotted by unrelated PRs. What could have landed in a day can take weeks to make it through the normal queue if, e.g., it touches half of Cargo.lock.
15:28larsbergThe scenario that's less relevant to Rust is for some of the shared autolander stuff we're doing with gecko where servo/servo is "sunk" into mozilla-central. They may need to backout a servo patch (they run tests AFTER landing), and since we don't give them direct push access to servo/servo, they need to at least be able to put that at the top of the queue
15:28larsbergand ensure nothing else lands and gets them even more broken :-)
15:28acrichtoaidanhs: you seem to know more about this than I, can you open an issue so I can cc it on the retry?
15:33aidanhs(not an urgent PR, but it's the best fix for now)
15:33acrichtoaidanhs: :(
15:33acrichtobut thanks!
16:39IcefozHmmm, so what's the functional difference between bors and homu?
16:40IcefozIt appears that rust is using one and servo+gfx is using the other? Or am I misunderstanding?
16:40edunhamIcefoz: they are different names for the same tool
16:40edunhamthe bot account who handles Rust stuff is a Homu instance with the username bors
16:40edunhamthe bot account who handles Servo stuff is a Homu instance with the name bors-servo
16:41edunhamhistorically, Bors was the original autolander and he ran on a cron job rather than listening to github's webhooks. This didn't scale, and a contributor wrote Homu as a replacement.
16:41IcefozOh, well then.
16:42IcefozThe old one just kept the name bors, then. All right!
16:42edunhamthe practical differences between Rust and Servo's homu instances is that one may be slightly behind the other if a new change has landed to, and they have different config files so different lists of people have r+
16:48steveklabniklarsberg: yeah, that makes sense
16:49steveklabniklarsberg: as always, sometimes things that are bad in one context make perfect sense in another
17:05frewsxcvanyone know what's going on here
17:06frewsxcvthat's from
17:09nagisafrewsxcv: I remember this happening a few times
17:09nagisanot sure what the fix was
17:10aidanhssee discussion above about removing the git caches
20:03simulacrumacrichto: Posted and
20:03simulacrumI think that's the first time I've seen infinite removed files.
20:05simulacrumInterestingly the other PR doesn't show infinity.
22:07aidanhser...bors doesn't seem to be running frewsxcv's rollup on appveyor
22:08aidanhsand the yellow blob next to the bors merge commit on does not list appveyor
22:08aidanhsam I missing something?
22:10aidanhsI think the travis build will be finishing in the next half hour, so I guess we'll see if bors merges
22:11aidanhsyup can't see it on there either
22:12aidanhshow exciting, I guess we get to see how reliable our CI process really is by whether bors merges or blows up :D
22:49frewsxcvaidanhs: oh, just saw your messages here
22:49frewsxcvlooks like travis finished but bors hasn't merged anything
22:49frewsxcvso....i guess bors is doing the right thing here
22:49aidanhsfrewsxcv: it's not finished yet
22:50aidanhsjust finishing off the last osx build
22:51frewsxcveverytime i load travis, it immediately kills my CPU and memory
22:51frewsxcvi wish they had a lite version of their site
22:51aidanhsI'm slowly working on that
22:53frewsxcvwell, now it's done
22:53aidanhsand no merge
22:53aidanhsgood to know
22:54frewsxcvi'm going to do a beta rollup, since those are more important right now
20 Apr 2017
No messages
Last message: 6 days and 1 hour ago