**** BEGIN LOGGING AT Thu May 02 02:59:57 2019 May 02 06:42:37 good morning May 02 17:51:41 JPEW, why do you use Yocto from inside a docker container, if I may ask ? May 02 17:52:38 kroon: 2 reasons: 1. Our developer environments are too disparate to get consistently reproducible builds otherwise May 02 17:53:12 2. We have to build Yocto releases all the way back to 2.2, and be able to reproduce them for the foreseeable future May 02 17:55:30 I see May 02 17:55:37 So for each release we have a container for the corresponding Ubuntu release that upstream recommends using, and we store that in the salt mines with our code when we release May 02 17:55:51 yeah, makes sense May 02 17:58:38 +1 May 02 18:04:52 kroon: If you are curious, we did open source our docker wrappers: https://github.com/garmin/pyrex They aren't *quite* where I'd like them to be... I really want to provide pre-build docker images instead of making everyone build locally May 02 18:06:05 There is also a lot that doesn't work, since we only implemented parts that pertained to our workflow (for example, we don't use devtool, so it's probably broken) May 02 18:09:17 JPEW, thanks, i'll give it a whirl, i've never used docker before May 02 18:10:34 kroon: I recommend an extensive reading of the README. It's *supposed* to be easy to setup, but it has some quirks. Also, feedback is welcome May 02 18:15:04 JPEW: did you try to optimize the build time at all? We use docker to build as well and while it works well, there was a mystery performance regression of 20+ a while ago that we never fixed/understood. May 02 18:15:34 vmeson: You mean the build time of bitbake itself? May 02 18:15:42 vmeson: I've never noticed a difference May 02 18:16:00 The build time in a docker image compared to native. May 02 18:16:28 vmeson: If there is a difference, it hasn't been enough for me to notice May 02 18:16:50 I should get some real recent numbers, ignore me for now. May 02 18:17:24 With all the SPECTRE issues being 'fixed' on buidld hosts, other build performance regressions haven't been as important. May 02 18:18:04 vmeson: hah. Well, to be fair I pretty much never build outside of docker these days, so maybe my builds are a lot slower than the could be May 02 18:18:18 As long as I get ~ 3000 builds/day from the cluster and the error rate is close to 2%, I'm happy. :) May 02 18:18:43 s/close to/under/ May 02 18:19:56 For the last 12 hrs, 'only' 18 of 1491 builds failed. May 02 18:20:49 and all the failures were on older releases so that's someone else's problem unless they really get out of whack! May 02 18:23:14 and while I'm chatty, 'only' 15 of 7732 run-time tests failed. May 02 18:29:27 3000 builds a day is impressive May 02 18:33:40 It's not Google scale but it's a start. It's just brute force with ~ 100 decent machines working on 6 YP releases. **** BEGIN LOGGING AT Thu May 02 20:01:16 2019 **** ENDING LOGGING AT Fri May 03 02:59:57 2019