| 15 Apr 2025 |
 connor (burnt/out) (UTC-8) | Jonas Chevalier: do you have any metrics on what load over time is like for the builders? | 20:49:32 |
 Gaétan Lepage | It would be intersting to know indeed.
From my personal experience, (I glance over the load of our 3 community builders regularly), I would say that the "average" LAV is quite low.
This doesn't say too much as when a build is occuring, they can very quickly be overloaded.
Some nix mechanism to prevent that (such as the work started by emily) would be very helpful.
Ideally a SLURM-like scheduler with a karma score would be the best way to share limited computing resources between several users. | 20:54:42 |
| Gaétan Lepage | * It would be intersting to know indeed.
From my personal experience, (I glance over the load of our 3 community builders regularly), I would say that the "average" LAV is quite low.
I.e., they are unused most of the time.
This doesn't say too much as when a build is occuring, they can very quickly be overloaded.
Some nix mechanism to prevent that (such as the work started by emily) would be very helpful.
Ideally a SLURM-like scheduler with a karma score would be the best way to share limited computing resources between several users. | 20:55:05 |
 emily | https://prometheus.nix-community.org/query?g0.expr=system_load1%7Bhost%3D%7E%22build01%7Cbuild05%7Cdarwin01%22%7D&g0.show_tree=0&g0.tab=graph&g0.range_input=2d&g0.res_type=auto&g0.res_density=medium&g0.display_mode=lines&g0.show_exemplars=0
build01 is build-box.nix-community.org (x86_64-linux)
build05 is aarch64-build-box.nix-community.org (aarch64-linux)
darwin01 is darwin-build-box.nix-community.org (aarch64-darwin) | 21:03:37 |
| connor (burnt/out) (UTC-8) | On a separate note, I was able to transfer cuda-legacy to https://github.com/nix-community/cuda-legacy, but now I can't edit the About to note myself and SomeoneSerge (UTC+U[-12,12]) are maintainers -- who can give me perms for the repo? | 21:13:16 |
 zowoq | Done. | 22:01:53 |
| 16 Apr 2025 |
|  Alexandru Tocar joined the room. | 11:48:32 |
| 18 Apr 2025 |
|  @cirnolovetech:matrix.org joined the room. | 01:10:37 |
|  lukegb (he/him) joined the room. | 17:29:35 |
| lukegb (he/him) | Is there any interest in me migrating https://github.com/lukegb/buildcatrust into nix-community? It's used for building the cacerts package in nixpkgs. | 17:36:32 |
|  @luxzi:matrix.org changed their profile picture. | 22:45:34 |
| zowoq | lukegb (he/him): I've sent you an invite to join the org. | 22:47:04 |
| 19 Apr 2025 |
|  Friday joined the room. | 00:37:05 |
|  @awwpotato:envs.net joined the room. | 03:58:22 |
| connor (burnt/out) (UTC-8) | Jonas Chevalier: I’d like to sponsor a GPU (CUDA) instance for nix-community so we can verify runtime behavior of things like PyTorch. Any ideas how much that would cost (or if Hetzner provides a discount for open source projects)? | 05:17:41 |
 hexa | https://www.hetzner.com/dedicated-rootserver/matrix-gpu | 09:45:50 |
| hexa | 221 USD/mo 🫣 | 09:46:53 |
|  jopejoe1 (4094@39c3) changed their display name from jopejoe1 to jopejoe1 (4094@eh22). | 12:59:48 |
|  @mig4ng:beeper.com left the room. | 16:18:18 |
|  @accelbread:matrix.org left the room. | 20:31:13 |
 emily | Grimmauld (any/all): are those git-of-theseus processes on the AArch64 box doing anything? it seems like they're using 0% CPU but a huge amount of memory | 21:53:35 |
| emily | (no worries if so, I just wasn't sure if they were leftovers are not :P) | 21:55:36 |
 Grimmauld (any/all) | In reply to @emilazy:matrix.org
Grimmauld (any/all): are those git-of-theseus processes on the AArch64 box doing anything? it seems like they're using 0% CPU but a huge amount of memory They are indexing nixpkgs, wanted to poke it a bit. The index takes 10h, is super slow, so I went and threw just more cores at it | 21:56:47 |
| Grimmauld (any/all) | If you need the memory for something productive you can kill them, but if they survive till tomorrow I'll have a nice graph :) | 21:57:41 |
| emily | I don't think I could kill them without abusing trusted-users status to escalate to root :P | 22:01:08 |
| emily | but I can run the eval on another builder | 22:01:13 |
| emily | (was going to do it locally except apparently CI eval is just broken on Darwin…) | 22:01:27 |
| Grimmauld (any/all) | Oh I can also drop to fewer processes, I believe its memory bandwidth limited | 22:02:02 |
| emily | the readme says https://github.com/src-d/hercules is significantly faster fwiw | 22:03:58 |
| Grimmauld (any/all) | I might have to put that into the PR that is packaging git-of-theseus | 22:05:36 |
