I’d prefer that we only move things to nightly that will not cause a tree closure in the event of failure. e.g., performance tests, valgrind runs, fuzzing, etc.
If we also put CSS / layout tests or other things we decide are “too slow” onto nightly builders, we’ll have the worst of both worlds - you still have to wait through a CI queue to land a fix, but we can also have tree closures, need sheriffs, and have to back out patches. I dealt with that since the 90s and, like most things I lived through in the 90s, would prefer never to experience it again :-) Up until now, we’ve managed with less than 1/10 of the hardware spend of the Rust team by carefully tweaking our infra. Now that our test and compile load is growing, we can use some of the budget (which we did allocate for 2015) to see if we can make it through 2015 until the TaskCluster folks get OSX support online and we can look to moving onto something a bit less ad-hoc than our massively redundant buildbot. That said, we’re always going to be gated on the Rust compile, especially for release. While the 45 minute OSX build appears to be a hardware issue, even at 20 minutes if we double the size of the Servo Rust codebase without reducing the sequential dependency chain and assume wistfully that Rust build times merely increase linearly with code size, we’re going to have a much bigger problem. Sorry for the long queue times! The requirement to switch to release builds because of the need to optimize ipc-channels and the Cargo limitation around that (https://github.com/rust-lang/cargo/issues/1359) put us into this situation months before we’d been expecting to hit it. - Lars On Aug 7, 2015, at 5:52 AM, Manish Goregaokar <[email protected]> wrote: I think we should be removing mac3 builds as well. I'll repeat what I proposed in IRC: We can instead have a "nightly" set of builds, for stuff like this which would be too much to run every build. This can include testing release builds, mac3, and future valgrind/android tests. If something breaks, we can bisect if necessary. -Manish Goregaokar On Fri, Aug 7, 2015 at 3:10 PM, Ms2ger <[email protected]> wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Hi all, > > The last mac3 run (<http://build.servo.org/builders/mac3/builds/829>) > took 1 hrs, 13 mins, 57 secs, of which it spent 45 minutes compiling. > As long as we have test-css running on linux (where css and wpt > together don't make to 30 minutes), does it make sense to have this > builder nearly double our turnaround time? > > Ms2ger > -----BEGIN PGP SIGNATURE----- > > iQEcBAEBAgAGBQJVxH0KAAoJEOXgvIL+s8n2d2wH/RsgXBnAaZA1W4Jw6bTp0kqL > pf0cl1Sjc56GK7vSd1C3BxnPqXFWuGJryuTJd7PuPElFfWiGBPlqp+ZRrsujVA23 > zQvKjR8SurGev3Q1tdNClVAiJo+mtLlVzHx6Vptk1SFzhl232MYWci+jMspTq8p6 > PNONLJ9B83SIRMAV5M9kvAQZKnw1yqthAEHd3OHPUKSyRxO5AX9kcPeKyLNvN+qK > Te13YXa9TeWB/8hXo//t0yNs5YV72mJ8exaIEiSVA2zkvQUZArOO5YEj7NfkT3BK > ugXNdyQLs/iQF8uiaaaqYmBThuiH6HJt09w8npvOYVE9hqvuuB0ir2SpLMS7NR4= > =SALA > -----END PGP SIGNATURE----- > _______________________________________________ > dev-servo mailing list > [email protected] > https://lists.mozilla.org/listinfo/dev-servo > _______________________________________________ dev-servo mailing list [email protected] https://lists.mozilla.org/listinfo/dev-servo _______________________________________________ dev-servo mailing list [email protected] https://lists.mozilla.org/listinfo/dev-servo
