| 14 Oct 2021 |
 moritz.hedtke | Wouldn't it be possible to run a build manually (with nix-build) on one of the idle ones (and probably disable it in hydra). I could image that would actually be faster. Although I don't know how it would get in the cache then or if there is some other flaw in that | 12:03:57 |
| moritz.hedtke | Investigating what's from is of course way more useful. Also this wouldn't update the channels in the end I assume and if hydra fails before building (which it seems to me) you couldn't even rerun a build there then | 12:07:30 |
| moritz.hedtke | * Investigating what's wrong is of course way more useful. Also this wouldn't update the channels in the end I assume and if hydra fails before building (which it seems to me) you couldn't even rerun a build there then | 12:07:48 |
 toonn | Things like stdenv updates require full rebuilds. That'd be ~28k packages to build manually. Somewhat less because you only need to manually start leaf builds but still. | 12:09:50 |
 lukegb (he/him) | At that point you have a human doing the Hydra scheduler's job, and the time would be better spent fixing the scheduler, imo | 12:10:52 |
 K900 | In reply to @lukegb:zxcvbnm.ninja
At the current rate, we're going to have a 21.11 release without x86_64-darwin builds, I'm afraid Is there a fixed deadline for 21.11? | 12:11:55 |
| toonn | Nov 26th is feature freeze I believe. | 12:12:12 |
| K900 | Yeah but like the actual release | 12:12:21 |
| toonn | If it's not in by then you can't really expect significant changes to still be accepted though. | 12:12:41 |
| K900 | I'm a random nobody so my opinion is irrelevant | 12:12:48 |
| K900 | But I'd rather delay the release than release without cached binaries for the second biggest platform | 12:13:04 |
| toonn | That's not really a great option without a solution in sight though. | 12:14:19 |
| toonn | Have many uncached expressions already made it into the unstable channel? | 12:14:39 |
| K900 | Lots, it seems | 12:14:49 |
| toonn | Because these are the kinds of things that would usually hold back channel bumps. | 12:15:08 |
| toonn | Which channel are you on? | 12:15:16 |
| K900 | nixpkgs-unstable | 12:15:24 |
| K900 | It looks like the CDN doesn't update when there are uncached expressions | 12:15:36 |
| K900 | But Github does | 12:15:38 |
| K900 | And I'm using flakes so I'm getting whatever is on Github | 12:15:44 |
| toonn | Ah, you're not actually using channels. You should pin your inputs to avoid things like this. Tracking master like that can always cause cache misses. | 12:16:34 |
| K900 | I'm not tracking master, I'm tracking the nixpkgs-unstable branch | 12:16:52 |
| K900 | And it's all pinned anyway cause flakes | 12:16:59 |
| toonn | That's not much different though. | 12:17:03 |
| K900 | Can you actually tell nix flake to track the CDN? | 12:17:23 |
| lukegb (he/him) | Nah, the nixpkgs-unstable *branch* should follow the channel | 12:17:24 |
| K900 | All the examples I've seen just use github:nixos/nixpkgs | 12:17:32 |
| K900 | Which is how I ended up here | 12:17:42 |
| lukegb (he/him) | The branch is, iirc, updated by the same infrastructure that updates the channel definitions anyway | 12:17:59 |
| toonn | Ah, there's no release-unstable? I know the branch naming isn't intuitive with the stable branches. | 12:18:04 |
