| 28 Feb 2024 |
 Alyssa Ross | Could we have higher darwin timeouts for staging PRs, or something? OfBorg Darwin is basically completely useless on staging PRs, because it can never get past LLVM, which takes >3600 seconds to build. | 14:17:09 |
 hexa | I think we're more affected by the sheer lack of build capacity for aarch64-darwin than its timeouts | 14:19:39 |
| hexa | but yeah, I've seen it often fail on llvm for … basically anything staging | 14:20:11 |
| Alyssa Ross | well, currently we're wasting a lot of that capacity on LLVM builds that are never going to finish | 14:20:39 |
| hexa | true | 14:20:46 |
| Alyssa Ross | so if we can't get more capacity, maybe we should just turn OfBorg Darwin off for staging | 14:21:03 |
| Alyssa Ross | since it's just wasting more than an hour of build time per PR, and is basically guaranteed not to produce any useful result | 14:21:31 |
 adamcstephens | there was some discussion above about moving to all aarch64 builders and running x86 on those instead, but i don't think any changes were made | 14:21:32 |
| adamcstephens | (not that that helps this specific timeout, but potentially it helps overall capacity) | 14:23:00 |
 Lily Foster | yeah increasing the timeout may help that problem for sure. we could just try it and see | 14:23:22 |
| hexa | 5x x86_64-linux, 1x aarch64-linux, 2x aarch64-darwin, 3x x86_64-darwin | 14:23:24 |
| hexa | *
- 5x x86_64-linux,
- 1x aarch64-linux,
- 2x aarch64-darwin,
- 3x x86_64-darwin
| 14:23:46 |
| hexa | *
- 5 x86_64-linux,
- 1 aarch64-linux,
- 2 aarch64-darwin,
- 3 x86_64-darwin
| 14:23:51 |
| adamcstephens | wow, we run all the aarch64-linux on one machine! and yet we never wait on it | 14:23:55 |
| Alyssa Ross | It's usually the fastest, even, IME | 14:24:09 |
| hexa | that is aarch64.nixos.community | 14:24:12 |
| hexa | fwiw | 14:24:13 |
| hexa | the community builder | 14:24:18 |
 7c6f434c | It has cores, lots of cores. | 14:24:20 |
| hexa | ampere altra q80-30 | 14:24:22 |
| Lily Foster | yeah the 1x for aarch64-linux is an absolute monster | 14:24:33 |
| hexa | the x86_64-linux machines co host the evaluators | 14:24:36 |
| Lily Foster | and actually runs like 20 ofborg consumers at once | 14:24:46 |
| hexa | 3 evaluators per host, 15 in parallel | 14:24:49 |
| Alyssa Ross | My latest OfBorg LLVM build for a staging PR on aarch46-darwin timed out with Ninja showing 2533/4188 | 14:25:05 |
| Alyssa Ross | but there's also the test suite after that, which takes a while | 14:25:11 |
| hexa | so 3h? | 14:25:17 |
| Alyssa Ross | so we might need to go to 2.5 or 3 hours | 14:25:17 |
| hexa | 
Download image.png | 14:25:49 |
| hexa | green is the queue for aarch64-darwin | 14:25:54 |
