| 3 Oct 2025 |
 Yureka (she/her) | I think the CACHIX_TOKEN is also passed directly into all CI runs on the projects (not on PRs though)? | 11:39:44 |
| Yureka (she/her) | 
Download foo.png | 11:40:58 |
| Yureka (she/her) |
This secret overrides an organization secret
| 11:41:08 |
|  hash joined the room. | 15:36:34 |
 hexa | I hope y'all progress on the self-hosted cache plan | 22:33:53 |
| hexa | and then ideally one cache per project | 22:34:06 |
|  moleksiak joined the room. | 23:16:58 |
| 4 Oct 2025 |
 zowoq | I guess this also means separate builders for each project with a further separation for untrusted builds (PRs from forks)? | 00:42:53 |
|  aleksana 🏳️⚧️ (force me to bed after 18:00 UTC) changed their profile picture. | 08:41:03 |
| aleksana 🏳️⚧️ (force me to bed after 18:00 UTC) changed their profile picture. | 08:41:27 |
|  dramforever joined the room. | 12:28:10 |
| dramforever | hi, i'm trying to help unblock the current staging-next -> master, and i need somewhere to test darwin stuff. would appreciate if someone would make this happen: https://github.com/nix-community/infra/pull/1992 | 12:53:52 |
| 5 Oct 2025 |
| hexa | Right, that makes it complicated fast | 14:34:54 |
| 6 Oct 2025 |
 Matt Sturgeon | How do we force buildbot to rebuild a merge queue group?
We had an intermittent failure on the PR's build which was solved by clicking rebuild.
IIUC, the same thing happened when we enqueued the PR; the merge group's build failed.
Now each time we queue the PR we get the cached failure, and we can't trigger a rebuild because GitHub deletes the merge group ref when it dequeues it.
(This PR: https://github.com/nix-community/nixvim/pull/3724) | 08:46:56 |
| Matt Sturgeon | Thanks @zowoq:matrix.org | 09:10:50 |
 Mic92 | @mattsturg:matrix.org: if you go to the eval build step, there is a restart button after github login | 09:28:45 |
| Mic92 | Latest buildbot-nix will also cache use the old evaluation, so it should only start failed evaluations | 09:29:49 |
| Mic92 | I should probably improve the message on cached failures regarding that. I have them actually disable in my installations, because they can only be restarted by repo owners and not contributera | 09:31:37 |
| Mic92 | * I should probably improve the message on cached failures regarding that. I have them actually disable in my installations, because they can only be restarted by repo owners and not contributers | 09:31:51 |
 SomeoneSerge (back on matrix) | Related: now trying to asses how much of hydra+builders+cache can be split up into microvms | 09:58:43 |
| SomeoneSerge (back on matrix) | One shared castore, multiple snix-stores | 09:59:44 |
| Matt Sturgeon | In reply to @joerg:thalheim.io
@mattsturg:matrix.org: if you go to the eval build step, there is a restart button after github login When I tried that for a merge group, I got:
fatal: couldn't find remote ref gh-readonly-queue/nixos-25.05/pr-3724-a30decbd5fc231e84dfefeb75bc7f57d8167726c
Because GitHub cleans up the merge group ref when it is dequeued, which happens as soon as it receives the failed status check. | 10:45:20 |
| Matt Sturgeon | (maybe buildbot should fetch the actual git object instead of fetching by reference to avoid this failure mode?) | 10:46:22 |
| Mic92 | In reply to @mattsturg:matrix.org
(maybe buildbot should fetch the actual git object instead of fetching by reference to avoid this failure mode?) Fetching by git commit is not a supported feature of the protocol | 11:02:44 |
| Mic92 | However we should not cache fetch errors in the first place | 11:03:21 |
| Mic92 | Can you open an issue | 11:03:30 |
| Mic92 | ? | 11:03:33 |
| Matt Sturgeon | In reply to @joerg:thalheim.io
Fetching by git commit is not a supported feature of the protocol You can do it via the git CLI git fetch <remote> <commit> (so I assume the git protocol supports it) and also via flakeref (if you know the commit id, at least: nix build github:NixOS/nixpkgs/<commit>). Which protocol does buildbot use to fetch? | 11:07:06 |
| Matt Sturgeon | In reply to @joerg:thalheim.io
However we should not cache fetch errors in the first place I haven't checked yet if that error caused cached failures; I saw it in a rebuild attempt | 11:07:40 |
| Mic92 | Redacted or Malformed Event | 11:08:25 |
