nix-config/modules/nixos/services/grafana/default.nix

{
  config,
  lib,
  namespace,
  ...
}:
with lib;
let
  name = "grafana";
  cfg = config.${namespace}.services.${name};

  giteaPort = config.${namespace}.services.gitea.port;
  resticPort = config.${namespace}.services.restic.port;
  nextcloudPort = config.${namespace}.services.nextcloud.port;

  grafanaConfig = lib.${namespace}.mkModule {
    inherit config name;
    description = "grafana";
    options = { };
    moduleConfig = {
      services = {
        prometheus = {
          enable = true;
          exporters = {
            node = {
              enable = true;
              enabledCollectors = [
                "filesystem"
                "diskstats"
                "meminfo"
                "cpu"
                "systemd"
                "processes"
              ];
              extraFlags = [
                "--collector.filesystem.mount-points-exclude=^/(dev|proc|sys|run)($|/)"
              ];
            };

            libvirt = {
              enable = false;
              openFirewall = true;
            };

            nut = {
              enable = true;
              openFirewall = true;
              passwordPath = config.sops.secrets."jallen-nas/ups_password".path;
              nutUser = upsUser;
            };

            # PostgreSQL — runs as the local postgres superuser via peer auth
            # (Unix socket, no password required).
            postgres = {
              enable = true;
              runAsLocalSuperUser = true;
            };

            # Redis — single exporter instance covering all four Redis servers
            # via the multi-target scrape pattern (/scrape?target=<addr>).
            # The exporter needs AF_INET to reach TCP Redis instances.
            redis = {
              enable = true;
              # No fixed --redis.addr: multi-target mode uses ?target= param.
            };

            # MariaDB — runs as the mysql OS user so it can connect via Unix
            # socket without a password.
            mysqld = {
              enable = true;
              runAsLocalSuperUser = true;
            };

            # Nextcloud — authenticates with the admin account.
            # passwordFile must be readable by the prometheus-nextcloud-exporter
            # user; sops mode 0440 + group keys covers that.
            nextcloud = {
              enable = true;
              url = "http://localhost:${toString nextcloudPort}";
              username = "mjallen";
              passwordFile = config.sops.secrets."jallen-nas/nextcloud/adminpassword".path;
            };
          };

          scrapeConfigs = [
            # ── System ──────────────────────────────────────────────────────────
            {
              job_name = "node";
              static_configs = [
                {
                  targets = [ "localhost:${toString config.services.prometheus.exporters.node.port}" ];
                }
              ];
            }

            # ── UPS (NUT) ────────────────────────────────────────────────────────
            {
              job_name = "nut";
              static_configs = [
                {
                  targets = [ "localhost:${toString config.services.prometheus.exporters.nut.port}" ];
                }
              ];
            }

            # ── Databases ────────────────────────────────────────────────────────
            {
              job_name = "postgres";
              static_configs = [
                {
                  targets = [ "localhost:${toString config.services.prometheus.exporters.postgres.port}" ];
                }
              ];
            }
            {
              # Redis multi-target: one exporter, four Redis instances.
              # The redis_exporter's /scrape?target= endpoint proxies each target
              # so a single exporter process covers all servers.
              job_name = "redis";
              metrics_path = "/scrape";
              static_configs = [
                {
                  targets = [
                    "redis://localhost:6379" # authentik
                    "redis://localhost:6363" # ccache
                    "redis://localhost:6380" # manyfold
                    "redis://localhost:6381" # onlyoffice
                  ];
                }
              ];
              relabel_configs = [
                {
                  source_labels = [ "__address__" ];
                  target_label = "__param_target";
                }
                {
                  source_labels = [ "__param_target" ];
                  target_label = "instance";
                }
                {
                  target_label = "__address__";
                  replacement = "localhost:${toString config.services.prometheus.exporters.redis.port}";
                }
              ];
            }
            {
              job_name = "mysqld";
              static_configs = [
                {
                  targets = [ "localhost:${toString config.services.prometheus.exporters.mysqld.port}" ];
                }
              ];
            }

            # ── Application services ─────────────────────────────────────────────
            {
              # Caddy exposes its built-in Prometheus endpoint on port 2019.
              job_name = "caddy";
              static_configs = [
                {
                  targets = [ "localhost:2019" ];
                }
              ];
            }
            {
              # Gitea's /metrics endpoint is protected by a Bearer token.
              job_name = "gitea";
              metrics_path = "/metrics";
              bearer_token_file = config.sops.secrets."jallen-nas/gitea/metrics-key".path;
              static_configs = [
                {
                  targets = [ "localhost:${toString giteaPort}" ];
                }
              ];
            }
            {
              # restic REST server exposes Prometheus metrics at /metrics.
              job_name = "restic";
              metrics_path = "/metrics";
              static_configs = [
                {
                  targets = [ "localhost:${toString resticPort}" ];
                }
              ];
            }
            {
              job_name = "nextcloud";
              static_configs = [
                {
                  targets = [ "localhost:${toString config.services.prometheus.exporters.nextcloud.port}" ];
                }
              ];
            }
          ];
        };

        grafana = {
          enable = true;
          settings = {
            server = {
              http_port = cfg.port;
              http_addr = "0.0.0.0";
            };
          };

          dataDir = "${cfg.configDir}/grafana";

          provision = {
            enable = true;
            datasources.settings.datasources = [
              {
                name = "Prometheus";
                type = "prometheus";
                access = "proxy";
                url = "http://localhost:${toString config.services.prometheus.port}";
              }
            ];
          };
        };
      };

      # The redis exporter needs AF_INET to reach TCP Redis instances.
      # The default systemd hardening only allows AF_UNIX.
      systemd.services.prometheus-redis-exporter.serviceConfig.RestrictAddressFamilies = [
        "AF_UNIX"
        "AF_INET"
        "AF_INET6"
      ];
    };
  };

  upsUser = "nas-admin";
in
{
  imports = [ grafanaConfig ];
}