nixos-servers/services/monitoring/prometheus.nix

{ self, lib, pkgs, ... }:
let
  monLib = import ../../lib/monitoring.nix { inherit lib; };
  externalTargets = import ./external-targets.nix;

  nodeExporterTargets = monLib.generateNodeExporterTargets self externalTargets;
  autoScrapeConfigs = monLib.generateScrapeConfigs self externalTargets;

  # Script to fetch AppRole token for Prometheus to use when scraping OpenBao metrics
  fetchOpenbaoToken = pkgs.writeShellScript "fetch-openbao-token" ''
    set -euo pipefail

    VAULT_ADDR="https://vault01.home.2rjus.net:8200"
    APPROLE_DIR="/var/lib/vault/approle"
    OUTPUT_FILE="/run/secrets/prometheus/openbao-token"

    # Read AppRole credentials
    if [ ! -f "$APPROLE_DIR/role-id" ] || [ ! -f "$APPROLE_DIR/secret-id" ]; then
      echo "AppRole credentials not found at $APPROLE_DIR" >&2
      exit 1
    fi

    ROLE_ID=$(cat "$APPROLE_DIR/role-id")
    SECRET_ID=$(cat "$APPROLE_DIR/secret-id")

    # Authenticate to Vault
    AUTH_RESPONSE=$(${pkgs.curl}/bin/curl -sf -k -X POST \
      -d "{\"role_id\":\"$ROLE_ID\",\"secret_id\":\"$SECRET_ID\"}" \
      "$VAULT_ADDR/v1/auth/approle/login")

    # Extract token
    VAULT_TOKEN=$(echo "$AUTH_RESPONSE" | ${pkgs.jq}/bin/jq -r '.auth.client_token')
    if [ -z "$VAULT_TOKEN" ] || [ "$VAULT_TOKEN" = "null" ]; then
      echo "Failed to extract Vault token from response" >&2
      exit 1
    fi

    # Write token to file
    mkdir -p "$(dirname "$OUTPUT_FILE")"
    echo -n "$VAULT_TOKEN" > "$OUTPUT_FILE"
    chown prometheus:prometheus "$OUTPUT_FILE"
    chmod 0400 "$OUTPUT_FILE"

    echo "Successfully fetched OpenBao token"
  '';
in
{
  # Systemd service to fetch AppRole token for Prometheus OpenBao scraping
  # The token is used to authenticate when scraping /v1/sys/metrics
  systemd.services.prometheus-openbao-token = {
    description = "Fetch OpenBao token for Prometheus metrics scraping";
    after = [ "network-online.target" ];
    wants = [ "network-online.target" ];
    before = [ "prometheus.service" ];
    requiredBy = [ "prometheus.service" ];

    serviceConfig = {
      Type = "oneshot";
      ExecStart = fetchOpenbaoToken;
      RemainAfterExit = true;
    };
  };

  # Timer to periodically refresh the token (AppRole tokens have 1-hour TTL)
  systemd.timers.prometheus-openbao-token = {
    description = "Refresh OpenBao token for Prometheus";
    wantedBy = [ "timers.target" ];
    timerConfig = {
      OnBootSec = "5min";
      OnUnitActiveSec = "30min";
      RandomizedDelaySec = "5min";
    };
  };

  services.prometheus = {
    enable = true;
    # syntax-only check because we use external credential files (e.g., openbao-token)
    checkConfig = "syntax-only";
    alertmanager = {
      enable = true;
      configuration = {
        global = {
        };
        route = {
          receiver = "webhook_natstonotify";
          group_wait = "30s";
          group_interval = "5m";
          repeat_interval = "1h";
          group_by = [ "alertname" ];
        };
        receivers = [
          {
            name = "webhook_natstonotify";
            webhook_configs = [
              {
                url = "http://localhost:5001/alert";
              }
            ];
          }
        ];
      };
    };
    alertmanagers = [
      {
        static_configs = [
          {
            targets = [ "localhost:9093" ];
          }
        ];
      }
    ];

    retentionTime = "30d";
    globalConfig = {
      scrape_interval = "15s";
    };
    rules = [
      (builtins.readFile ./rules.yml)
    ];

    scrapeConfigs = [
      # Auto-generated node-exporter targets from flake hosts + external
      {
        job_name = "node-exporter";
        static_configs = [
          {
            targets = nodeExporterTargets;
          }
        ];
      }
      # Systemd exporter on all hosts (same targets, different port)
      {
        job_name = "systemd-exporter";
        static_configs = [
          {
            targets = map (t: builtins.replaceStrings [":9100"] [":9558"] t) nodeExporterTargets;
          }
        ];
      }
      # Local monitoring services (not auto-generated)
      {
        job_name = "prometheus";
        static_configs = [
          {
            targets = [ "localhost:9090" ];
          }
        ];
      }
      {
        job_name = "loki";
        static_configs = [
          {
            targets = [ "localhost:3100" ];
          }
        ];
      }
      {
        job_name = "grafana";
        static_configs = [
          {
            targets = [ "localhost:3000" ];
          }
        ];
      }
      {
        job_name = "alertmanager";
        static_configs = [
          {
            targets = [ "localhost:9093" ];
          }
        ];
      }
      {
        job_name = "pushgateway";
        honor_labels = true;
        static_configs = [
          {
            targets = [ "localhost:9091" ];
          }
        ];
      }
      {
        job_name = "labmon";
        static_configs = [
          {
            targets = [ "monitoring01.home.2rjus.net:9969" ];
          }
        ];
      }
      # TODO: nix-cache_caddy can't be auto-generated because the cert is issued
      # for nix-cache.home.2rjus.net (service CNAME), not nix-cache01 (hostname).
      # Consider adding a target override to homelab.monitoring.scrapeTargets.
      {
        job_name = "nix-cache_caddy";
        scheme = "https";
        static_configs = [
          {
            targets = [ "nix-cache.home.2rjus.net" ];
          }
        ];
      }
      # pve-exporter with complex relabel config
      {
        job_name = "pve-exporter";
        static_configs = [
          {
            targets = [ "10.69.12.75" ];
          }
        ];
        metrics_path = "/pve";
        params = {
          module = [ "default" ];
          cluster = [ "1" ];
          node = [ "1" ];
        };
        relabel_configs = [
          {
            source_labels = [ "__address__" ];
            target_label = "__param_target";
          }
          {
            source_labels = [ "__param_target" ];
            target_label = "instance";
          }
          {
            target_label = "__address__";
            replacement = "127.0.0.1:9221";
          }
        ];
      }
      # OpenBao metrics with bearer token auth
      {
        job_name = "openbao";
        scheme = "https";
        metrics_path = "/v1/sys/metrics";
        params = {
          format = [ "prometheus" ];
        };
        static_configs = [{
          targets = [ "vault01.home.2rjus.net:8200" ];
        }];
        authorization = {
          type = "Bearer";
          credentials_file = "/run/secrets/prometheus/openbao-token";
        };
      }
    ] ++ autoScrapeConfigs;

    pushgateway = {
      enable = true;
      web = {
        external-url = "https://pushgw.home.2rjus.net";
      };
    };
  };
}