| 4 Jan 2024 |
 Gaétan Lepage | Hello !
I am trying to use the buildbot for nix-community/nixvim.
I added the nix-community-buildbot tag to the repo.
Is there something else to do ?
It doesn't seem to be working (nixvim is not listed here) | 08:22:03 |
| Gaétan Lepage | NOTE: I just discovered that "Newly enabled repos are imported into buildbot twice a day, you can also ask the admins to trigger an import manually." | 08:30:17 |
| Gaétan Lepage | * NOTE: I just discovered that "Newly enabled repos are imported into buildbot twice a day, you can also ask the admins to trigger an import manually."
I will then wait for a few hours :) | 08:30:33 |
 zowoq | I triggered an import for you. | 08:31:47 |
| Gaétan Lepage | Thanks !!! | 08:31:52 |
| Gaétan Lepage | Our checks have a lot of outputs (more than 1000). Could that explain why buildbot is very slow compared to github actions ?
It looks like it processes all of the individual outputs individually instead of running a single time nix flake check. | 08:53:43 |
| Gaétan Lepage | https://buildbot.nix-community.org/#/builders/542/builds/1 | 08:53:47 |
|  lillecarl changed their display name from LilleCarl to LilleCarl (Salsa9). | 09:06:26 |
 Mic92 | This is a design tradeoff. It allows to have per-package logs / failures. For some flakes like disko, also nix flake check would go out of memory because there are too many nixos machines it would to evaluate. buildbot-nix splits evaluation and build | 09:50:02 |
| Mic92 | I haven't yet experimented with very large build matrixes yet. | 09:50:29 |
| Mic92 | It would be interesting to see how quickly rebuilds are done. | 09:52:30 |
| Mic92 | I would expect that many builds will be skipped and don't actually spawn any process. | 09:52:50 |
| Mic92 | It's good to see that the UI somewhat scales | 09:55:39 |
| Mic92 | Do you have numbers how fast github is able to build those 1000 derivations? | 09:56:47 |
| Gaétan Lepage | In reply to @joerg:thalheim.io
Do you have numbers how fast github is able to build those 1000 derivations? Recently, we switch to my own self-hosted github runner (Ryzen 5600X 6C/12T, 16GB) and it takes a few minutes at most (depends on what is already available on cachix) | 10:40:26 |
| Mic92 | Ok. but how long does a full rebuild takes? | 10:51:50 |
| Mic92 | There is also some caching done after a second build | 10:52:30 |
| Gaétan Lepage | It took 1 hour and 48 minutes | 11:13:58 |
| Mic92 | How long on your server? | 11:33:02 |
| Gaétan Lepage | <10min | 12:26:53 |
| Mic92 | You probably also didn't had to do any remote builds, which also seem a bottleneck | 12:36:39 |
| Gaétan Lepage | Yes, it looks like I could run nix flake check --all-systems locally even though I had no remote-builders | 12:43:00 |
| Mic92 | what are those module- packages? | 12:49:27 |
| Mic92 | It looks like nix-eval-jobs report those to be not cached. | 12:49:40 |
| Mic92 | Also cachix than says during upload they are. | 12:50:05 |
| Mic92 | mhm locally they are considered cached | 12:56:44 |
| Mic92 | Another bottleneck might be also that it update's the job status on github for every derivation. | 13:08:49 |
| Mic92 | All in all as of today, I haven't really optimized it for these types of workloads. Hydra might be still the better choice for large package sets. | 13:15:21 |
| 5 Jan 2024 |
|  obsqrprjkt joined the room. | 14:08:52 |
| 6 Jan 2024 |
|  eyJhb set a profile picture. | 14:00:35 |
