| 14 May 2024 |
 Qyriad | fear | 00:36:36 |
| Qyriad | I think I would rather try to convince infinisil of lua | 00:36:55 |
 raitobezarius | Back on the topic of CI and compute: I'd be wary of multiplying too many providers at the same time because it makes it hard to maintain them on the long run, I think, if we can have on-demand compute 24/7 for the four major platforms, that's really comfortable and then we can look into additional compute for special tasks if needed | 00:37:43 |
| raitobezarius | (thinking about things like Windows or BSD VMs but that's extremely long term) | 00:38:09 |
 samrose | This is what it one of the main builds of postgresql looks like on cirrus fwiw https://cirrus-ci.com/github/postgresql-cfbot/postgresql | 00:40:04 |
| raitobezarius | I don't see the defn of the CI tasks in the repo | 00:41:36 |
| samrose | In reply to @raitobezarius:matrix.org
I don't see the defn of the CI tasks in the repo it looks like they just mirror to github, and run their own git server | 00:43:21 |
| samrose | https://git.postgresql.org/gitweb/ | 00:43:29 |
| samrose | although some of the cirrus build stuff can be seen at https://github.com/postgres/postgres | 00:44:08 |
| samrose | this page says https://cirrus-ci.org/guide/linux/
For arm_container instances Cirrus CI uses a EKS cluster of Graviton2 instances running in AWS.
| 00:44:47 |
| raitobezarius | OK so container based and it doesn't seem to support dynamic CI targets like we do right now | 00:46:15 |
| raitobezarius | Everything has to be spelled in advance | 00:46:21 |
| samrose | raitobezarius: what are the dynamic targets looking like? | 00:47:13 |
| raitobezarius | Unsure if it's worth the trouble compared to just obtaining an Altra Q80 tbh | 00:47:15 |
| samrose | yeah it may not be | 00:47:23 |
| raitobezarius | In reply to @samrose:matrix.org
raitobezarius: what are the dynamic targets looking like? Just nix-eval-jobs on the flake.nix and generate a task for each hydraJobs | 00:47:43 |
| samrose | ah ok | 00:47:58 |
| samrose | you can't do that in a container in any case can you? | 00:48:22 |
| raitobezarius | Orthogonal, you can do it in a container, but it doesn't seem like Cirrus has a phase where you tell it "ok spawn me N containers and make them do this and that" | 00:49:07 |
| raitobezarius | Thus it's necessarily serial inside of that container or you statically list all your targets as you change them | 00:49:28 |
| samrose | I see | 00:49:30 |
| raitobezarius | Plus the container means you go from clean Nix store and you need to redownload stuff | 00:50:00 |
| raitobezarius | Or you need to build some hot container, push it to registry and load it in the CI | 00:50:14 |
| raitobezarius | Whereas current model share the Nix store and GC it as it go | 00:50:28 |
| samrose | so the optimal way would be to have the actual hardware, and be running nix-eval-jobs and just let nix do it's thing + retain the store basically like the way many people have used hydra in the past | 00:51:12 |
| raitobezarius | I'd say so, right now, an improvement would be to go to something like Hydra New Generation | 00:51:56 |
| samrose | it doesn't sound like it would be a lot of fun to maintain building this in containers the way you describe it | 00:52:00 |
| raitobezarius | In reply to @samrose:matrix.org
it doesn't sound like it would be a lot of fun to maintain building this in containers the way you describe it Hahahahahahaha :-) | 00:52:13 |
| samrose | I mean, not that I think you are wrong at all | 00:52:25 |
| raitobezarius | Oh, those are my 2 cents | 00:52:43 |
