NixOS's systemd abstraction doesn't work with systemd template units

[grahamc]

Describe the bug

A clear and concise description of what the bug is.

Steps To Reproduce

Steps to reproduce the behavior:

Create a template unit, like:

{
  systemd.services."foo@" = {
    scriptArgs = "%i";
    script = ''
      echo "$1"
    '';
  };

  systemd.services."foo@example" = {};
}

Expected behavior

Starting foo@example should run the foo@example template service.

Actual behavior

Starting a template unit which is not defined by NixOS works fine:

# systemctl start foo@bar
# journalctl -fu foo@bar
-- Journal begins at Wed 2021-08-11 22:25:17 EDT. --
Aug 24 10:01:35 scruffy systemd[1]: Started foo@bar.service.
Aug 24 10:01:35 scruffy foo_-start[3297689]: bar
Aug 24 10:01:35 scruffy systemd[1]: foo@bar.service: Succeeded.

However, starting the foo@example service, defined by NixOS fails:

# systemctl status foo@example
● foo@example.service
     Loaded: bad-setting (Reason: Unit foo@example.service has a bad unit file setting.)
     Active: inactive (dead)

Aug 24 10:01:41 scruffy systemd[1]: foo@example.service: Service has no ExecStart=, ExecStop=, or SuccessAction=. Refusing.

Notify maintainers

Metadata

Please run nix-shell -p nix-info --run "nix-info -m" and paste the result.

• system: "x86_64-linux"
• host os: Linux 5.10.52, NixOS, 21.11.20210810.6f6ade6 (Porcupine)
• multi-user?: yes
• sandbox: yes
• version: nix-env (Nix) 2.4pre20210813_d581129
• nixpkgs: /nix/var/nix/profiles/per-user/root/channels/nixos

Maintainer information:

# a list of nixpkgs attributes affected by the problem
attribute:
# a list of nixos modules affected by the problem
module:

[grahamc]

Further, I added another service like:

{
  systemd.services."several-foos" = {
      requires = [ "foo@fizz" "foo@buzz" ]; 
      wantedBy = [ "default.target" ];
      script = ''
      
      '';
 }

and several-foos starts but foo@fizz and foo@buzz do not get started.

[vcunat]

I've been successfully generating a target. Re-tested now on 21.05:

{
    systemd.targets.kresd = { # configure units started by default
      wantedBy = [ "multi-user.target" ];
      wants = [ "kres-cache-gc.service" ]
        ++ map (i: "kresd@${toString i}.service") (range 1 cfg.instances);
    };
 }

[grahamc]

Is the kresd@.service unit file provided by kresd the package, or by a NixOS service?

[vcunat]

Provided by package and modified by service.

[grahamc]

I think the issue is unique to units which are created totally by the module system.

[vcunat]

As a workaround you should be able to define a pseudo-package that defines the service (i.e. creates the unit file "manually").

[rnhmjoj]

Duplicate of #80933?

[pennae]

using instances of systemd services create by the module system in OnFailure= clauses does work on recent nixos-unstable (c21ba4f), as does the snippet from #135557 (comment) after suffixing .service to each required unit.

this does feel like it's behaving correctly: referencing templated units with parameters elsewhere invokes them as expected, and defining them through the module system defines them as though they hadn't existed before. maybe we should have a systemd.dropIn.<name> to explicitly configure overrides?

[roberth]

I think we should first model Exec* correctly before we revise are replace .script. #154123

[ck3d]

PR #186314 is now merged into master, which fixes this issue. A working example:

{
  systemd.services."foo@" = {
    scriptArgs = "%i";
    script = ''
      echo "$1"
    '';
  };

  systemd.services."foo@example" = {
    overrideStrategy = "asDropin";
  };
}

[con-f-use]

I don't think that example works anymore and we might be looking at a regression here. As in if I copy&paste this the foo@example job is not started on system boot but rather displayed in systemctl status foo@example.service as dead. The corresponding service file does not have an exec line or anything other than environment definitions:

$ cat /etc/systemd/system/foo@example.service.d/overrides.conf 
[Unit]

[Service]
Environment="LOCALE_ARCHIVE=/nix/store/9dij3pl4dkdxmxhasjw1pa9hzqv4rjlp-glibc-locales-2.39-52/lib/locale/locale-archive"
Environment="PATH=/nix/store/ysqx2xfzygv2rxl7nxnw48276z5ckppn-coreutils-9.5/bin:/nix/store/36rvynxwln7iz0qq3k1v3r1mna8bma8s-fi"
Environment="TZDIR=/nix/store/jyh52p2cxrjn8r4ywdv2am5pjkj1xcqa-tzdata-2024a/share/zoneinfo"

---

Reproduction:

# build with: nix-build <this file>
# then run: result/bin/run-nixos-vm
let
  configuration = 
    { pkgs, lib, config, ... }:
    {
      systemd.services."foo@" = {
        scriptArgs = "%i";
        script = ''
          echo "$1"
        '';
      };

      systemd.services."foo@example" = {
        overrideStrategy = "asDropin";
      };

      services.getty.autologinUser = "root";

      system.stateVersion = config.system.nixos.release;

      virtualisation.vmVariant = {
        virtualisation = {
          qemu.options = [ "-vga virtio" ];
          diskImage = null; # "./nixos.qcow2";
          mountHostNixStore = true;
          writableStoreUseTmpfs = false;
        };
      };

      boot.loader.grub.device = "nodev";
      fileSystems."/" = { };
    }
  ;

  commit = "a046c1202e11b62cbede5385ba64908feb7bfac4";
  source = builtins.fetchTarball {
    url = "https://github.com/NixOS/nixpkgs/archive/${commit}.tar.gz";
  };
in
(import "${source}/nixos/lib/eval-config.nix" {
  system = "x86_64-linux";
  modules = [ configuration ];
}).config.virtualisation.vmVariant.system.build.vm

@ck3d @grahamc

[ck3d]

The override.conf should not have an exec setting, the the template service defines exec. Also the status dead looks as expected.
I see no degradation.