111 lines
4.1 KiB
Nix
111 lines
4.1 KiB
Nix
{ config, lib, pkgs, ... }:
|
|
|
|
let
|
|
cfg = config.services.nix-agent;
|
|
extraArgs =
|
|
map (arg: "--health-check-command ${lib.escapeShellArg arg}") cfg.healthCheckCommand
|
|
++ lib.optionals cfg.rollbackOnFailure [ "--rollback-on-failure" ]
|
|
++ lib.optionals cfg.apply [ "--apply" ];
|
|
renderedExtraArgs = lib.concatStringsSep " \\\n " extraArgs;
|
|
in
|
|
{
|
|
options.services.nix-agent = {
|
|
enable = lib.mkEnableOption "UltraCloud nix-agent service (host OS convergence only, consuming desired-system state published by deployer)";
|
|
|
|
chainfireEndpoint = lib.mkOption {
|
|
type = lib.types.str;
|
|
default = "http://127.0.0.1:2379";
|
|
description = "ChainFire endpoint consumed by nix-agent for desired-system data; nix-agent does not schedule native services.";
|
|
};
|
|
|
|
clusterNamespace = lib.mkOption {
|
|
type = lib.types.str;
|
|
default = "ultracloud";
|
|
description = "Cluster namespace prefix";
|
|
};
|
|
|
|
clusterId = lib.mkOption {
|
|
type = lib.types.str;
|
|
description = "Cluster ID reconciled by nix-agent";
|
|
};
|
|
|
|
nodeId = lib.mkOption {
|
|
type = lib.types.str;
|
|
default = config.networking.hostName;
|
|
description = "Node ID represented by this agent";
|
|
};
|
|
|
|
flakeRoot = lib.mkOption {
|
|
type = lib.types.str;
|
|
default = "/etc/nixos";
|
|
description = "Flake root used to build target nixosConfigurations";
|
|
};
|
|
|
|
intervalSecs = lib.mkOption {
|
|
type = lib.types.int;
|
|
default = 30;
|
|
description = "Polling interval in seconds";
|
|
};
|
|
|
|
switchAction = lib.mkOption {
|
|
type = lib.types.enum [ "switch" "test" "boot" "dry-activate" ];
|
|
default = "switch";
|
|
description = "switch-to-configuration action executed after building the target system";
|
|
};
|
|
|
|
healthCheckCommand = lib.mkOption {
|
|
type = lib.types.listOf lib.types.str;
|
|
default = [ ];
|
|
description = "Command vector executed after activation to verify node health. Entries are argv items, not a shell snippet; a non-zero exit triggers rollback when rollbackOnFailure is enabled.";
|
|
};
|
|
|
|
rollbackOnFailure = lib.mkOption {
|
|
type = lib.types.bool;
|
|
default = true;
|
|
description = "Roll back to the previous system if the post-activation health check fails, leaving observed status as rolled-back instead of keeping the rejected target active.";
|
|
};
|
|
|
|
apply = lib.mkOption {
|
|
type = lib.types.bool;
|
|
default = true;
|
|
description = "Apply desired NixOS system state on the node; runtime process placement remains the node-agent and fleet-scheduler path.";
|
|
};
|
|
|
|
package = lib.mkOption {
|
|
type = lib.types.package;
|
|
default = pkgs.nix-agent or (throw "nix-agent package not found");
|
|
description = "Package to use for nix-agent";
|
|
};
|
|
};
|
|
|
|
config = lib.mkIf cfg.enable {
|
|
systemd.services.nix-agent = {
|
|
description = "UltraCloud Nix Agent (host OS reconcile only)";
|
|
wantedBy = [ "multi-user.target" ];
|
|
after = [ "network-online.target" ];
|
|
wants = [ "network-online.target" ];
|
|
path = [ config.system.path ];
|
|
|
|
serviceConfig = {
|
|
Type = "simple";
|
|
Restart = "on-failure";
|
|
RestartSec = "5s";
|
|
# The agent can invoke switch-to-configuration on its own host. Keep
|
|
# that child process alive when systemd stops the agent during the
|
|
# switch transaction so activation can finish and restart services in
|
|
# the new generation.
|
|
KillMode = "process";
|
|
ExecStart = ''
|
|
${cfg.package}/bin/nix-agent \
|
|
--chainfire-endpoint ${lib.escapeShellArg cfg.chainfireEndpoint} \
|
|
--cluster-namespace ${lib.escapeShellArg cfg.clusterNamespace} \
|
|
--cluster-id ${lib.escapeShellArg cfg.clusterId} \
|
|
--node-id ${lib.escapeShellArg cfg.nodeId} \
|
|
--flake-root ${lib.escapeShellArg cfg.flakeRoot} \
|
|
--interval-secs ${toString cfg.intervalSecs} \
|
|
--switch-action ${lib.escapeShellArg cfg.switchAction}${lib.optionalString (renderedExtraArgs != "") " \\\n ${renderedExtraArgs}"}
|
|
'';
|
|
};
|
|
};
|
|
};
|
|
}
|