| 30 Aug 2025 |
 Vladimír Čunát | * Ah, it seems like /root/.cache/nix/binary-cache-v6.sqlite* is the most important, at least for my workflows (judging by timestamps on the files). | 08:17:47 |
| Vladimír Čunát | nixos-25.05 should update in about one hour. | 09:05:53 |
| 31 Aug 2025 |
|  sinan changed their profile picture. | 16:25:25 |
| 1 Sep 2025 |
 sterni (he/him) | This build seems to lock up indefinitely on sleepy-brown while doing some local networking stuff: https://hydra.nixos.org/build/306255307 Can't reproduce this problem locally (with Lix). Last success was on a different machine 5 days earlier. | 13:50:41 |
 hexa | restarted once more | 13:54:07 |
| hexa | the host is currently idle, so maybe that helps | 13:54:27 |
| hexa | build@elated-minsky.builder.nixos.org x86_64-linux,i686-linux /var/lib/hydra/queue-runner/.ssh/id_ed25519 32 4 kvm,nixos-test,benchmark - -
build@sleepy-brown.builder.nixos.org x86_64-linux,i686-linux /var/lib/hydra/queue-runner/.ssh/id_ed25519 4 4 kvm,nixos-test,big-parallel big-parallel -
build@sleepy-brown.builder.nixos.org x86_64-linux,i686-linux /var/lib/hydra/queue-runner/.ssh/id_ed25519 8 4 kvm,nixos-test,big-parallel - -
| 13:55:09 |
| hexa | fwiw | 13:55:10 |
| hexa | normal builds can run on both builders | 13:55:30 |
| sterni (he/him) | okay ty. didn't want to potentially take up a slot for 3h again if I forget to cancel the build before asking if anything relevant changed in the last days | 13:55:30 |
| hexa | big-parallel only on sleepy-brown | 13:55:35 |
| sterni (he/him) | sounds plausible that the test is poorly implemented and just broken on busy machines, this is fairly common ime, though I don't see lockups very often | 13:56:28 |
| hexa | no relevant config changes were made to the x86_64 builders | 13:56:35 |
| hexa | sterni: looks done 😄 | 14:00:41 |
| sterni (he/him) | third time's the charm | 14:08:14 |
|  Lun changed their display name from lun to Lun. | 15:55:22 |
| Lun | will hydra be sad if master gets hit with composable_kernel rebuilds eg via https://github.com/NixOS/nixpkgs/pull/436350 ? | 16:42:23 |
 emily | that PR needs to get down to 0 rebuilds | 16:42:37 |
| emily | so no worries. I'll comment there | 16:42:56 |
| Lun | I may have got too many crimes in-tree for that to be practical 😨 | 16:44:49 |
| emily | you can always get a refactor like this down to 0 rebuilds if you put your mind to it :) | 16:45:20 |
| emily | (and it's the only way to be reasonably reassured you're not regressing anything) | 16:45:29 |
| emily | fwiw I don't think individual huge builds are anywhere near as bad as many builds, because of the bottlenecks around the queue runner scheduling jobs etc., as opposed to just taking up slots on big-parallel machines | 16:45:36 |
| emily | but maybe for ones this huge best to avoid it on master, dunno | 16:45:45 |
 Tristan Ross | The problem is splicing is changing things for ROCm from the looks of it. I tried looking at how to do it for ROCm's LLVM but the new splicing stuff has changed the behavior of overriding things. | 16:51:04 |
| emily | let's move this to https://matrix.to/#/!kxOJEqURGkuOHTRRQB:matrix.org?via=matrix.org&via=nixos.dev&via=tchncs.de perhaps? | 16:52:13 |
| Lun | It'd be good to get confirmation on the original question about whether master getting CK rebuilds is ok outside the context of this specific PR - I have a few things targeting staging right now that would be better to target master if the answer is that it's fine. | 16:56:51 |
| emily | my Hydra intuition says that it's not that bad but that obviously 1 × few but big rebuilds beats 5 × few but big rebuilds if you can avoid it. (e.g., merging PRs in batches would achieve the same result.) but I'll defer to the experts | 17:02:08 |
| hexa | yes, we can a eat a few fat builds on master | 17:06:49 |
| hexa | that's why we have big-parallel slots | 17:06:56 |
