| 4 Jan 2025 |
emily | we're already seeing e.g. not being able to test the Firefox build on AArch64 because the community builder is gone | 00:40:38 |
emily | that's quite painful as aarch64-linux is our second best supported platform after x86_64-linux currently | 00:40:54 |
liberodark | It's actually in my pipeline.
But these are very expensive servers. | 00:41:02 |
emily | (and increasingly commonly deployed in the cloud etc.) | 00:41:06 |
emily | yeah :( | 00:41:08 |
liberodark | Unfortunately I also do with the means I have at my disposal.
I think that to help on this I need a little time to negotiate this point.
I have noted that this was important for you.
But I try to respond to the needs 1 by 1.
The first being equinix. | 00:42:53 |
emily | FWIW Equinix also provided the community/ofborg AArch64 builders, so I think we currently have no plan for aarch64-linux pre-merge CI too | 00:43:53 |
emily | but yeah that's the hardest thing to solve of course | 00:43:59 |
liberodark | Indeed, it was also indicated to me. | 00:44:58 |
liberodark | And I agree with you that this is not easy to resolve but not impossible. I have some ideas for this. | 00:45:41 |
zowoq | Upgrading the x86 build box to an epyc 9454p would be nice. | 00:49:03 |
emily | I'm always in favour of more cores :) | 00:49:18 |
emily | but I also like testing on aarch64-linux because ~everything already gets built on x86_64-linux during development | 00:49:33 |
zowoq | Where would this machine be located? Who covers the hosting cost? | 01:05:57 |
liberodark | I propose the machine. It is hosted at tetaneutral. | 01:07:40 |
liberodark | Those who can, my contribution is on the hardware and support part for hosting as I do for other services or associations. | 01:08:58 |
liberodark | This reduces your costs compared to Hetzner, for example. | 01:11:56 |
liberodark | After you can give me an installation by nix-anywhere and I push it on the machine you have a dedicated fullstack v4 and v6 IP. | 01:12:51 |
liberodark | * After you can give me an installation by nix-anywhere (for example) and I push it on the machine you have a dedicated fullstack v4 and v6 IP. | 01:13:09 |
zowoq | Sorry, I not sure that I understand correctly.
We do need to cover the hosting cost?
Who has access to the machine if we need a manual reboot? | 01:48:31 |
zowoq | * Sorry, I'm not sure that I understand correctly.
We do need to cover the hosting cost?
Who has access to the machine if we need a manual reboot? | 01:51:09 |
tomberek | In reply to @liberodark:matrix.org In my case I propose to provide a machine for free to help. The hosting cost is 25€ per server. Since we have dedicated 10 Gb lines. Main costs are for electricity more than anything else. On my side, i am only proposing if it can help. This might also be helpful for infra or staging. Are you in touch with the right people? | 02:28:15 |
emily | AIUi the plan is for ofborg to use these machines | 02:28:28 |
emily | I don't think Hydra can realistically use machines that the infra team doesn't have total control over and that don't have KVM | 02:28:45 |
tomberek | CI is the most obvious. Nix-community might benefit. Or even dedicated access for a team, stdenv perhaps? | 02:30:23 |
emily | this is nix-community :) | 02:31:39 |
zowoq | Putting another machine towards ofborg or using it as an secondary builder (not replacing the current hetzner build box) may be the better option. Saving a bit of money would be nice but if it was the only x86 community builder not having kvm/ipmi access to it would probably be a recurring problem. | 03:03:19 |
liberodark | I am in contact with some person from nixos. | 08:49:56 |
liberodark | But for nixos-community I learned about the builder issue only yesterday. | 08:50:31 |
liberodark | I was advised to talk to zowoq. | 08:50:53 |