08:16 <aminechikhaoui> aanderse sorry just noticed your message, it isn't clear to me what's slow in the deployment, sounds like the build of the required derivations itself is slow. I guess it's not really NixOPS related ? even if you nix-build the derivations it would be slow ?

08:19 <aminechikhaoui> Raito_Bezarius I think hydra.nixos.org is deployed using flakes, https://github.com/NixOS/nixos-org-configurations/blob/master/delft/flake.nix#L8

08:20 <aminechikhaoui> iirc it's a matter of doing nixops create -d <depl> --flake <flake-uri>

12:02 <aanderse> aminechikhaoui: true, but because it's nixops there are many machines, so i thought you might have some experience... how large are you nixops networks compared to how fast they build?

13:15 <aminechikhaoui> aanderse because our builds are slow we build things on a hydra instance then use the resulting closure to deploy, so generally things built during the deploy are minimal. We've had issues before with large networks (50ish machines) when the output size of builds is large but that was an issue during the copy closure phase mostly.

13:17 <aanderse> aminechikhaoui: thanks. that information is useful. it at least tells me there isn't any magic nixops specific sauce i can throw on this to fix it... have to look at specific machines and improve there, or come up with solutions like you mention (overnight deploys, etc...)

16:13 <Raito_Bezarius> aanderse: wouldn't flake eval caching help in your case?

16:14 <aanderse> Raito_Bezarius: i don't know, i don't really know much about flakes

16:15 <Raito_Bezarius> If you grep the logs on NixOps evaluation questions, I think I have seen some discussions with clever and adisbladis on how to make it faster under certain conditions but it's on the Nix level (keyword: builtins.fork)

16:43 <aminechikhaoui> flakes would help with eval caching but my understanding is that this is just few builds being slow not evaluations ?