Hi! I've been using NixOS for some 6 years, and I feel pretty comfortable with it. But I've never made the leap to Flakes, and know very little about them. But I want to learn more.

I have large repository of configuration modules that I use to compose nixoses for several personal machines. I've also used nix at work to deploy several web services. For the web services, I'd typically create:

• A package.nix that defines how my source code is built into an executable;
• a service.nix that defines the systemd units and whatever config is needed to run it in a machine;
• a default.nix (containing only callPackage ./package.nix {} or something) for local builds and shells; and
• a deployment.nix (or production.nix / acceptance.nix) that has the environment and hardware specific things.

I deploy by simply using nixos-rebuild deployment.nix --target-host .... When I have multiple web-services in a single repository, the whole structure just moves into the corresponding sub-directories, and I call nixos-rebuild on each of 'em.

In the deployment pipeline I pin nixpkgs to a specific hash, but that doesn't guarantee that local builds and shells use the same version of nixpkgs and whatnot. Today, I'm looking into migrating one of these projects to Flakes, mainly to improve reproducibility.

I'm looking for some general guidelines as to how to go about this.

• Should I be creating multiple flakes? Like, one for the package and one for the service? Or should I combine everything into a single flake?
• In the monorepo case, should I have a single flake at the root that just defines multiple build outputs? I can imagine that that makes managing locks easier? Or should I stick with a flake per package / target host?
• How do I reuse packages? Do I still create derivations that I can import with callPackage or does Flakes replace it?

In reply to @avaq:matrix.org

Hi! I've been using NixOS for some 6 years, and I feel pretty comfortable with it. But I've never made the leap to Flakes, and know very little about them. But I want to learn more.

I have large repository of configuration modules that I use to compose nixoses for several personal machines. I've also used nix at work to deploy several web services. For the web services, I'd typically create:

• A package.nix that defines how my source code is built into an executable;
• a service.nix that defines the systemd units and whatever config is needed to run it in a machine;
• a default.nix (containing only callPackage ./package.nix {} or something) for local builds and shells; and
• a deployment.nix (or production.nix / acceptance.nix) that has the environment and hardware specific things.

I deploy by simply using nixos-rebuild deployment.nix --target-host .... When I have multiple web-services in a single repository, the whole structure just moves into the corresponding sub-directories, and I call nixos-rebuild on each of 'em.

In the deployment pipeline I pin nixpkgs to a specific hash, but that doesn't guarantee that local builds and shells use the same version of nixpkgs and whatnot. Today, I'm looking into migrating one of these projects to Flakes, mainly to improve reproducibility.

I'm looking for some general guidelines as to how to go about this.

• Should I be creating multiple flakes? Like, one for the package and one for the service? Or should I combine everything into a single flake?
• In the monorepo case, should I have a single flake at the root that just defines multiple build outputs? I can imagine that that makes managing locks easier? Or should I stick with a flake per package / target host?
• How do I reuse packages? Do I still create derivations that I can import with callPackage or does Flakes replace it?