A simple multi-profile Nix-flake deploy tool.
Go to file
2023-11-02 09:25:24 +01:00
.github/workflows [Chore] Run CI checks on 'pull_request' 2023-09-18 12:40:24 +02:00
.reuse chore: replace .license files with dep5 config 2022-01-13 15:38:18 -07:00
docs chore: replace .license files with dep5 config 2022-01-13 15:38:18 -07:00
examples fixup! [#210] Add activation script for darwin system and provide a usage example 2023-05-12 10:59:19 +08:00
LICENSES Add license information, reformat Nix files, clean up 2020-09-29 15:10:06 -07:00
src [Chore] fix error messages claiming to have rolled back when not actually doing so 2023-11-02 09:25:24 +01:00
.envrc Add .envrc and fix devShell to work with more tools 2021-08-04 00:06:51 -07:00
.gitignore fix: ignore direnv files 2021-08-26 13:49:35 -05:00
Cargo.lock [#201] Deduce profile directory during activation 2023-09-12 12:00:17 +02:00
Cargo.toml [#201] Deduce profile directory during activation 2023-09-12 12:00:17 +02:00
default.nix Add missing license information 2020-10-11 15:23:58 -07:00
flake.lock Automatically update flake.lock to the latest version (#152) 2022-12-29 18:48:32 +04:00
flake.nix Replace jsonschema-cli with check-jsonschema 2023-09-17 10:34:25 +02:00
interface.json Add missing newline to interface.json 2020-11-10 16:37:10 -07:00
LICENSE chore: link LICENSES/MPL-2.0.txt to root LICENSE 2022-01-13 15:38:18 -07:00
README.md Merge pull request #231 from serokell/rvem/#201-dont-hardcode-profile-directory 2023-09-12 12:15:07 +02:00
shell.nix Add reuse headers to shell.nix and .buildkite/pipeline.yml 2020-11-14 20:07:45 -07:00

deploy-rs logo


A Simple, multi-profile Nix-flake deploy tool.

Questions? Need help? Join us on Matrix: #deploy-rs:matrix.org

Usage

Basic usage: deploy [options] <flake>.

Using this method all profiles specified in the given <flake> will be deployed (taking into account the profilesOrder).

Optionally the flake can be constrained to deploy just a single node (my-flake#my-node) or a profile (my-flake#my-node.my-profile).

If your profile or node name has a . in it, simply wrap it in quotes, and the flake path in quotes (to avoid shell escaping), for example 'my-flake#"myserver.com".system'.

Any "extra" arguments will be passed into the Nix calls, so for instance to deploy an impure profile, you may use deploy . -- --impure (note the explicit flake path is necessary for doing this).

You can try out this tool easily with nix run:

  • nix run github:serokell/deploy-rs your-flake

If you want to deploy multiple flakes or a subset of profiles with one invocation, instead of calling deploy <flake> you can issue deploy --targets <flake> [<flake> ...] where <flake> is supposed to take the same format as discussed before.

Running in this mode, if any of the deploys fails, the deploy will be aborted and all successful deploys rolled back. --rollback-succeeded false can be used to override this behavior, otherwise the auto-rollback argument takes precedent.

If you require a signing key to push closures to your server, specify the path to it in the LOCAL_KEY environment variable.

Check out deploy --help for CLI flags! Remember to check there before making one-time changes to things like hostnames.

There is also an activate binary though this should be ignored, it is only used internally (on the deployed system) and for testing/hacking purposes.

Ideas

deploy-rs is a simple Rust program that will take a Nix flake and use it to deploy any of your defined profiles to your nodes. This is strongly based off of serokell/deploy, designed to replace it and expand upon it.

Multi-profile

This type of design (as opposed to more traditional tools like NixOps or morph) allows for lesser-privileged deployments, and the ability to update different things independently of each other. You can deploy any type of profile to any user, not just a NixOS profile to root.

Magic Rollback

There is a built-in feature to prevent you making changes that might render your machine unconnectable or unusuable, which works by connecting to the machine after profile activation to confirm the machine is still available, and instructing the target node to automatically roll back if it is not confirmed. If you do not disable magicRollback in your configuration (see later sections) or with the CLI flag, you will be unable to make changes to the system which will affect you connecting to it (changing SSH port, changing your IP, etc).

API

Overall usage

deploy-rs is designed to be used with Nix flakes (this currently requires an unstable version of Nix to work with). There is a Flake-less mode of operation which will automatically be used if your available Nix version does not support flakes, however you will likely want to use a flake anyway, just with flake-compat (see this wiki page for usage).

deploy-rs also outputs a lib attribute, with tools used to make your definitions simpler and safer, including deploy-rs.lib.${system}.activate (see later section "Profile"), and deploy-rs.lib.${system}.deployChecks which will let nix flake check ensure your deployment is defined correctly.

There are full working deploy-rs Nix expressions in the examples folder, and there is a JSON schema here which is used internally by the deployChecks mentioned above to validate your expressions.

A basic example of a flake that works with deploy-rs and deploys a simple NixOS configuration could look like this

{
  description = "Deployment for my server cluster";

  # For accessing `deploy-rs`'s utility Nix functions
  inputs.deploy-rs.url = "github:serokell/deploy-rs";

  outputs = { self, nixpkgs, deploy-rs }: {
    nixosConfigurations.some-random-system = nixpkgs.lib.nixosSystem {
      system = "x86_64-linux";
      modules = [ ./some-random-system/configuration.nix ];
    };

    deploy.nodes.some-random-system.profiles.system = {
        user = "root";
        path = deploy-rs.lib.x86_64-linux.activate.nixos self.nixosConfigurations.some-random-system;
    };

    # This is highly advised, and will prevent many possible mistakes
    checks = builtins.mapAttrs (system: deployLib: deployLib.deployChecks self.deploy) deploy-rs.lib;
  };
}

In the above configuration, deploy-rs is built from the flake, not from nixpkgs. To take advantage of the nixpkgs binary cache, the deploy-rs package can be overwritten in an overlay:

{
  # ...
  outputs = { self, nixpkgs, deploy-rs }: let
    system = "x86_64-linux";
    # Unmodified nixpkgs
    pkgs = import nixpkgs { inherit system; };
    # nixpkgs with deploy-rs overlay but force the nixpkgs package
    deployPkgs = import nixpkgs {
      inherit system;
      overlays = [
        deploy-rs.overlay
        (self: super: { deploy-rs = { inherit (pkgs) deploy-rs; lib = super.deploy-rs.lib; }; })
      ];
    };
  in {
    # ...
    deploy.nodes.some-random-system.profiles.system = {
        user = "root";
        path = deployPkgs.deploy-rs.lib.activate.nixos self.nixosConfigurations.some-random-system;
    };
  };
}

Profile

This is the core of how deploy-rs was designed, any number of these can run on a node, as any user (see further down for specifying user information). If you want to mimic the behaviour of traditional tools like NixOps or Morph, try just defining one profile called system, as root, containing a nixosSystem, and you can even similarly use home-manager on any non-privileged user.

{
  # A derivation containing your required software, and a script to activate it in `${path}/deploy-rs-activate`
  # For ease of use, `deploy-rs` provides a function to easily add the required activation script to any derivation
  # Both the working directory and `$PROFILE` will point to `profilePath`
  path = deploy-rs.lib.x86_64-linux.activate.custom pkgs.hello "./bin/hello";

  # An optional path to where your profile should be installed to, this is useful if you want to use a common profile name across multiple users, but would have conflicts in your node's profile list.
  # This will default to `"/nix/var/nix/profiles/system` if `user` is `root` and profile name is `system`,
  # `/nix/var/nix/profiles/per-user/root/$PROFILE_NAME` if profile name is different.
  # For non-root profiles will default to /nix/var/nix/profiles/per-user/$USER/$PROFILE_NAME if `/nix/var/nix/profiles/per-user/$USER` already exists,
  # and `${XDG_STATE_HOME:-$HOME/.local/state}/nix/profiles/$PROFILE_NAME` otherwise.
  profilePath = "/home/someuser/.local/state/nix/profiles/someprofile";

  # ...generic options... (see lower section)
}

Node

This defines a single node/server, and the profiles you intend it to run.

{
  # The hostname of your server. Can be overridden at invocation time with a flag.
  hostname = "my.server.gov";

  # An optional list containing the order you want profiles to be deployed.
  # This will take effect whenever you run `deploy` without specifying a profile, causing it to deploy every profile automatically.
  # Any profiles not in this list will still be deployed (in an arbitrary order) after those which are listed
  profilesOrder = [ "something" "system" ];

  profiles = {
    # Definition format shown above
    system = {};
    something = {};
  };

  # ...generic options... (see lower section)
}

Deploy

This is the top level attribute containing all of the options for this tool

{
  nodes = {
    # Definition format shown above
    my-node = {};
    another-node = {};
  };

  # ...generic options... (see lower section)
}

Generic options

This is a set of options that can be put in any of the above definitions, with the priority being profile > node > deploy

{
  # This is the user that deploy-rs will use when connecting.
  # This will default to your own username if not specified anywhere
  sshUser = "admin";

  # This is the user that the profile will be deployed to (will use sudo if not the same as above).
  # If `sshUser` is specified, this will be the default (though it will _not_ default to your own username)
  user = "root";

  # Which sudo command to use. Must accept at least two arguments:
  # the user name to execute commands as and the rest is the command to execute
  # This will default to "sudo -u" if not specified anywhere.
  sudo = "doas -u";

  # This is an optional list of arguments that will be passed to SSH.
  sshOpts = [ "-p" "2121" ];

  # Fast connection to the node. If this is true, copy the whole closure instead of letting the node substitute.
  # This defaults to `false`
  fastConnection = false;

  # If the previous profile should be re-activated if activation fails.
  # This defaults to `true`
  autoRollback = true;

  # See the earlier section about Magic Rollback for more information.
  # This defaults to `true`
  magicRollback = true;

  # The path which deploy-rs will use for temporary files, this is currently only used by `magicRollback` to create an inotify watcher in for confirmations
  # If not specified, this will default to `/tmp`
  # (if `magicRollback` is in use, this _must_ be writable by `user`)
  tempPath = "/home/someuser/.deploy-rs";

  # Build the derivation on the target system. 
  # Will also fetch all external dependencies from the target system's substituters.
  # This default to `false`
  remoteBuild = true;
}

About Serokell

deploy-rs is maintained and funded with ❤️ by Serokell. The names and logo for Serokell are trademark of Serokell OÜ.

We love open source software! See our other projects or hire us to design, develop and grow your idea!