modules/prometheus: refactor

.sops.yaml

@@ -2,12 +2,13 @@ keys:
   - &xin age1uw059wcwfvd9xuj0hpqzqpeg7qemecspjrsatg37wc7rs2pumfdsgken0c
   - &host-calcite age1ytwfqfeez3dqtazyjltn7mznccwx3ua8djhned7n8mxqhw4p6e5s97skfa
   - &host-raspite age1nugzw24upk8pz5lyz2z89qk8se4gpcsg3ypcs58nykncr56sevrsm8qpvj
-  - &host-sgp-00 age13s6rwd3wjk2x5wkn69tdczhl3l5d7mfmlv90efsv4q67jne43qss9tcakx
   - &host-la-00 age1fw2sqaa5s9c8ml6ncsexkj8ar4288387ju92ytjys4awf9aw6smqqz94dh
   - &host-massicot age1jle2auermhswqtehww9gqada8car5aczrx43ztzqf9wtcld0sfmqzaecta
   - &host-weilite age17r3fxfmt6hgwe984w4lds9u0cnkf5ttq8hnqt800ayfmx7t8t5gqjddyml
   - &host-hk-00 age1p2dlc8gfgyrvtta6mty2pezjycn244gmvh456qd3wvkfwesp253qnwyta9
   - &host-fra-00 age18u4mqrhqkrpcytxfxfex6aeap04u38emhy6u4wrp5k62sz2vae4qm5jj7s
+  - &host-biotite age1v5h946jfke6ae8pcgz52mhj26cacqcpl9dmmrrkf37x55rnq2v3szqctvv
+  - &host-thorite age12ng08vjx5jde5ncqutwkd5vm4ygfwy33mzhzwe0lkxzglulgpqusc89r96
 creation_rules:
   - path_regex: machines/calcite/secrets.yaml
     key_groups:
@@ -24,19 +25,18 @@ creation_rules:
     - age:
       - *xin
       - *host-massicot
+  - path_regex: machines/thorite/secrets.yaml
+    key_groups:
+    - age:
+      - *xin
+      - *host-thorite
   - path_regex: machines/dolomite/secrets/secrets.yaml
     key_groups:
     - age:
       - *xin
-      - *host-sgp-00
       - *host-la-00
       - *host-hk-00
       - *host-fra-00
-  - path_regex: machines/dolomite/secrets/sgp-00.yaml
-    key_groups:
-    - age:
-      - *xin
-      - *host-sgp-00
   - path_regex: machines/dolomite/secrets/la-00.yaml
     key_groups:
     - age:
@@ -64,7 +64,6 @@ creation_rules:
       - *xin
       - *host-calcite
       - *host-raspite
-      - *host-sgp-00
       - *host-la-00
       - *host-hk-00
       - *host-massicot

flake.nix

@@ -60,8 +60,8 @@
   outputs =
     {
       self,
-      home-manager,
       nixpkgs,
+      home-manager,
       nixos-hardware,
       sops-nix,
       flake-utils,
@@ -84,6 +84,7 @@
       overlayModule =
         { ... }:
         {
+          _module.args.my-lib = import ./overlays/my-lib;
           nixpkgs.overlays = [
             editorOverlay
             (import ./overlays/add-pkgs.nix)
@@ -118,10 +119,6 @@
           ./machines/dolomite/bandwagon.nix
           ./machines/dolomite/common.nix
         ];
-        tok-00 = [
-          ./machines/dolomite/lightsail.nix
-          ./machines/dolomite/common.nix
-        ];
         fra-00 = [
           ./machines/dolomite/fra.nix
           ./machines/dolomite/common.nix
@@ -178,7 +175,6 @@
         };
     in
     {
-      nixpkgs = nixpkgs;
       nixosModules.default = {
         imports = [
           ./modules/nixos
@@ -190,7 +186,9 @@
       colmenaHive = colmena.lib.makeHive {
         meta = {
           # FIXME:
-          nixpkgs = import nixpkgs { system = "x86_64-linux"; };
+          nixpkgs = import nixpkgs {
+            system = "x86_64-linux";
+          };
         };
 
         massicot =
@@ -205,20 +203,6 @@
             ] ++ sharedColmenaModules;
           };
 
-        tok-00 =
-          { ... }:
-          {
-            imports = nodeNixosModules.tok-00 ++ sharedColmenaModules;
-            nixpkgs.system = "x86_64-linux";
-            networking.hostName = "tok-00";
-            system.stateVersion = "23.11";
-            deployment = {
-              targetHost = "video01.namely.icu";
-              buildOnTarget = false;
-              tags = [ "proxy" ];
-            };
-          };
-
         la-00 =
           { ... }:
           {
@@ -310,7 +294,6 @@
         osmium = mkNixos {
           hostname = "osmium";
         };
-
       } // self.colmenaHive.nodes;
 
     }

machines/biotite/default.nix

@@ -12,11 +12,12 @@
   networking.useNetworkd = true;
   systemd.network.enable = true;
   systemd.network.networks."10-wan" = {
-    matchConfig.MACAddress = "00:16:3e:0a:ec:45";
+    matchConfig.MACAddress = "b6:20:0d:9a:6c:34";
-    networkConfig.DHCP = "ipv4";
+    networkConfig = {
-    dhcpV4Config = {
+      DHCP = "ipv4";
-      UseDNS = true;
+      IPv6SendRA = true;
     };
+    address = [ "2a03:4000:4a:148::1/64" ];
   };
 
   commonSettings = {

machines/calcite/configuration.nix

@@ -368,8 +368,7 @@ in
   };
 
   custom.prometheus = {
-    enable = true;
+    exporters.node.enable = true;
-    exporters.blackbox.enable = true;
   };
 
   services.ollama = {

machines/dolomite/claw.nix

@@ -31,14 +31,7 @@
       fsType = "ext4";
     };
 
-    swapDevices = [ ];
+    networking.useNetworkd = true;
-
-    # Enables DHCP on each ethernet and wireless interface. In case of scripted networking
-    # (the default) this is the recommended approach. When using systemd-networkd it's
-    # still possible to use this option, but it's recommended to use it in conjunction
-    # with explicit per-interface declarations with `networking.interfaces.<interface>.useDHCP`.
-    # networking.useNetworkd = false;
-
     systemd.network.enable = true;
     systemd.network.networks."10-wan" = {
       matchConfig.MACAddress = "00:16:3e:0a:ec:45";
@@ -47,7 +40,6 @@
         UseDNS = true;
       };
     };
-    # networking.interfaces.eth0.useDHCP = lib.mkDefault true;
 
     nixpkgs.hostPlatform = lib.mkDefault "x86_64-linux";
   };

machines/dolomite/common.nix

@@ -21,12 +21,20 @@
         };
       };
     };
+    swapDevices = [
+      {
+        device = "/swapfile";
+        size = 2 * 1024;
+      }
+    ];
 
-    custom.prometheus = {
+    custom.prometheus.exporters = {
-      enable = lib.mkDefault true;
+      enable = true;
-      exporters.blackbox.enable = true;
+      node.enable = true;
     };
 
+    services.tailscale.enable = true;
+
     commonSettings = {
       auth.enable = true;
       proxyServer = {

machines/dolomite/fra.nix

@@ -52,6 +52,12 @@
     networkConfig = {
       DHCP = "no";
       Gateway = "185.217.108.1";
+      DNSSEC = true;
+      DNSOverTLS = true;
+      DNS = [
+        "8.8.8.8#dns.google"
+        "8.8.4.4#dns.google"
+      ];
     };
     address = [ "185.217.108.59/24" ];
   };

machines/massicot/default.nix

@@ -24,10 +24,6 @@
       hedgedoc_env = {
         owner = "hedgedoc";
       };
-      grafana_cloud_api = {
-        owner = "prometheus";
-        sopsFile = ../secrets.yaml;
-      };
       grafana_oauth_secret = {
         owner = "grafana";
       };
@@ -62,6 +58,8 @@
     hostName = "massicot";
   };
 
+  services.tailscale.enable = true;
+
   commonSettings = {
     auth.enable = true;
     nix = {

machines/massicot/networking.nix

@@ -3,8 +3,10 @@
   networking.useNetworkd = true;
   systemd.network.networks."10-wan" = {
     matchConfig.MACAddress = "96:00:02:68:7d:2d";
-    networkConfig.DHCP = "ipv4";
+    networkConfig = {
-    networkConfig.Gateway = "fe80::1";
+      DHCP = "ipv4";
+      Gateway = "fe80::1";
+    };
     address = [
       "2a01:4f8:c17:345f::3/64"
     ];

machines/massicot/services.nix

@@ -43,10 +43,14 @@ in
     environmentFile = config.sops.secrets.hedgedoc_env.path;
   };
 
-  custom.prometheus = {
+  custom.prometheus.exporters = {
     enable = true;
-    exporters.blackbox.enable = true;
+    blackbox = {
-    exporters.miniflux.enable = true;
+      enable = true;
+    };
+    node = {
+      enable = true;
+    };
   };
 
   security.acme = {

machines/thorite/default.nix

@@ -1,29 +1,41 @@
-{ ... }:
 {
   imports = [
     ./hardware-configurations.nix
+    ./monitoring.nix
   ];
 
-  networking.hostName = "thorite";
+  config = {
-  networking.useNetworkd = true;
+    networking.hostName = "thorite";
-  systemd.network.enable = true;
+    networking.useNetworkd = true;
-  systemd.network.networks."10-wan" = {
+    systemd.network.enable = true;
-    matchConfig.MACAddress = "00:51:d3:21:f3:28";
+    systemd.network.networks."10-wan" = {
-    networkConfig = {
+      matchConfig.MACAddress = "00:51:d3:21:f3:28";
-      DHCP = "no";
+      networkConfig = {
-      Gateway = "23.165.200.1";
+        DHCP = "no";
+        Gateway = "23.165.200.1";
+        DNSSEC = true;
+        DNSOverTLS = true;
+        DNS = [
+          "8.8.8.8#dns.google"
+          "8.8.4.4#dns.google"
+        ];
+      };
+      address = [ "23.165.200.99/24" ];
     };
-    address = [ "23.165.200.99/24" ];
+
+    networking.firewall.allowedTCPPorts = [
+      80
+      443
+    ];
+
+    commonSettings = {
+      auth.enable = true;
+      autoupgrade.enable = true;
+    };
+
+    nixpkgs.system = "x86_64-linux";
+    system.stateVersion = "24.11";
+
+    users.users.root.hashedPassword = "$y$j9T$NToEZWJBONjSgRnMd9Ur9/$o6n7a9b8eUILQz4d37oiHCCVnDJ8hZTZt.c.37zFfU.";
   };
-
-  nixpkgs.system = "x86_64-linux";
-
-  system.stateVersion = "24.11";
-
-  commonSettings = {
-    auth.enable = true;
-    autoupgrade.enable = true;
-  };
-
-  users.users.root.hashedPassword = "$y$j9T$NToEZWJBONjSgRnMd9Ur9/$o6n7a9b8eUILQz4d37oiHCCVnDJ8hZTZt.c.37zFfU.";
 }

machines/thorite/monitoring.nix

@@ -0,0 +1,100 @@
+{ config, my-lib, ... }:
+with my-lib;
+{
+  config = {
+    sops = {
+      defaultSopsFile = ./secrets.yaml;
+      age.sshKeyPaths = [ "/etc/ssh/ssh_host_ed25519_key" ];
+      secrets = {
+        "grafana/oauth_secret" = {
+          owner = "grafana";
+        };
+      };
+    };
+
+    custom.monitoring = {
+      grafana.enable = true;
+    };
+
+    services.caddy.virtualHosts."https://grafana.xinyang.life".extraConfig =
+      with config.services.grafana.settings.server; ''
+        reverse_proxy http://${http_addr}:${toString http_port}
+      '';
+
+    custom.prometheus = {
+      enable = true;
+      exporters = {
+        enable = true;
+        blackbox.enable = true;
+        node.enable = true;
+      };
+      ruleModules = (mkCaddyRules [ { host = "thorite"; } ]) ++ (mkNodeRules [ { host = "thorite"; } ]);
+    };
+
+    services.prometheus.scrapeConfigs =
+      let
+        probeList = [
+          "la-00.video.namely.icu:8080"
+          "fre-00.video.namely.icu:8080"
+          "hk-00.video.namely.icu:8080"
+          "49.13.13.122:443"
+          "45.142.178.32:22"
+          "home.xinyang.life:8000"
+        ];
+      in
+      (mkScrapes [
+        {
+          name = "immich";
+          scheme = "http";
+          address = "weilite.coho-tet.ts.net";
+          port = 8082;
+        }
+        {
+          name = "gotosocial";
+          address = "xinyang.life";
+        }
+        {
+          name = "miniflux";
+          address = "rss.xinyang.life";
+        }
+        {
+          name = "ntfy";
+          address = "ntfy.xinyang.life";
+        }
+        {
+          name = "grafana-eu";
+          address = "grafana.xinyang.life";
+        }
+      ])
+      ++ (mkCaddyScrapes [
+        { address = "thorite.coho-tet.ts.net"; }
+      ])
+      ++ (mkNodeScrapes [
+        { address = "thorite.coho-tet.ts.net"; }
+        { address = "massicot.coho-tet.ts.net"; }
+        { address = "weilite.coho-tet.ts.net"; }
+        { address = "hk-00.coho-tet.ts.net"; }
+        { address = "la-00.coho-tet.ts.net"; }
+        { address = "fra-00.coho-tet.ts.net"; }
+      ])
+      ++ (mkBlackboxScrapes [
+        {
+          hostAddress = "thorite.coho-tet.ts.net";
+          targetAddresses = probeList;
+        }
+        {
+          hostAddress = "massicot.coho-tet.ts.net";
+          targetAddresses = probeList;
+        }
+        {
+          hostAddress = "weilite.coho-tet.ts.net";
+          targetAddresses = [
+            "la-00.video.namely.icu:8080"
+            "fre-00.video.namely.icu:8080"
+            "hk-00.video.namely.icu:8080"
+          ];
+        }
+      ]);
+
+  };
+}

machines/thorite/secrets.yaml

@@ -0,0 +1,31 @@
+grafana:
+    oauth_secret: ENC[AES256_GCM,data:angZR3sl8vGcbAXyKFBvCSm+YhF5OooCcxRiSxR2zBoXMz5wv5/uMJFynwOTRVI6,iv:hVpOlM89lNbK6AsGf4Is/tLv3xPfg/XdtA8vuEK52L8=,tag:zCER+IdRnTcG2WHQ/AhxZA==,type:str]
+sops:
+    kms: []
+    gcp_kms: []
+    azure_kv: []
+    hc_vault: []
+    age:
+        - recipient: age1uw059wcwfvd9xuj0hpqzqpeg7qemecspjrsatg37wc7rs2pumfdsgken0c
+          enc: |
+            -----BEGIN AGE ENCRYPTED FILE-----
+            YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBoTXRtTlRES3V4MGhZaGdr
+            aXJ4UFNDT0Nrb0ZuWEkxUEFDU2orbzNBSVhVCkh2VitqMGwwOVdhMFJIeWU1eTgw
+            UVdxY0tLVDJNVnRnQmMyS0FPYS9LVmMKLS0tIEZaMTdIMU5SQUkxL2NFK2Jtbm9v
+            YVR3RHpDR3F2aFlCWGd5TjNOV2p4YzgK8OKpwcvTK/0j+kQCo0+8n6sQ5Pu9t9xZ
+            lPWeUGk1BudsyCqgIZWF5iXfu1pJnYq1XEAM0ttJl402xKeqIovM0Q==
+            -----END AGE ENCRYPTED FILE-----
+        - recipient: age12ng08vjx5jde5ncqutwkd5vm4ygfwy33mzhzwe0lkxzglulgpqusc89r96
+          enc: |
+            -----BEGIN AGE ENCRYPTED FILE-----
+            YWdlLWVuY3J5cHRpb24ub3JnL3YxCi0+IFgyNTUxOSBIYk1NTlhsYW8xbFppNTBE
+            WlJmNzhnclBoVENXa1cvcHY5NGdRZVAzV0FJClpsTHpTeG9CK3J2ZFEreG1BTWpG
+            WjdaYzlLQnU0LzJLSDBZZ2pvOWdvSEkKLS0tIExRT0p1aCttZG5MMW14emJmRk5w
+            M2pqMUJoMGlBZnpBaVBUTFFRZUMzb2sKrlWy26Cv55/8XQEl9hee8P29uj582sIx
+            mUjaYE0U2qOP9bklXUQyyzQjfkBLWTLc1PTX9BjqOOsqXwkRQIYppA==
+            -----END AGE ENCRYPTED FILE-----
+    lastmodified: "2024-11-28T17:02:03Z"
+    mac: ENC[AES256_GCM,data:14FOUXuKP+8+sad1UlhBW37fWzmutpyn6d4q2qKtBiOyT5ivHunFHJfHrtX83X2fLDmUfiD42bXf+rYfdtKzVUmQ6vutCUQk+Hal8NElhjcq5Ns5kT4VZRKG7/ya9+eNEEkajtq/7OFEM5KOQKTKjyOBqBq/AdYQ+ni9r45c1sM=,iv:WrdWSfrZrGalZO4WGk3JpgACY7W0odt3vP+pRkMXHfA=,tag:jeRBfR2QYjLBylOLHxU3hQ==,type:str]
+    pgp: []
+    unencrypted_suffix: _unencrypted
+    version: 3.9.1

machines/weilite/default.nix

@@ -16,7 +16,6 @@
     networking.hostName = "weilite";
     commonSettings = {
       auth.enable = true;
-      autoupgrade.enable = true;
       nix = {
         enable = true;
         enableMirrors = true;
@@ -61,8 +60,14 @@
       };
     };
 
-    custom.prometheus = {
+    custom.prometheus.exporters = {
       enable = true;
+      blackbox = {
+        enable = true;
+      };
+      node = {
+        enable = true;
+      };
     };
 
     systemd.mounts = [

modules/nixos/prometheus/blackbox.nix

@@ -1,93 +0,0 @@
-{
-  config,
-  lib,
-  pkgs,
-  ...
-}:
-let
-  cfg = config.custom.prometheus;
-in
-{
-  config = lib.mkIf (cfg.enable && cfg.exporters.blackbox.enable) {
-    services.prometheus.exporters.blackbox = {
-      enable = true;
-      listenAddress = "127.0.0.1";
-      configFile = pkgs.writeText "blackbox.config.yaml" (
-        lib.generators.toYAML { } {
-          modules = {
-            tcp4_connect = {
-              prober = "tcp";
-              tcp = {
-                ip_protocol_fallback = false;
-                preferred_ip_protocol = "ip4";
-                tls = false;
-              };
-              timeout = "15s";
-            };
-          };
-        }
-      );
-    };
-
-    services.prometheus.scrapeConfigs = [
-      {
-        job_name = "blackbox";
-        scrape_interval = "1m";
-        metrics_path = "/probe";
-        params = {
-          module = [ "tcp4_connect" ];
-        };
-        static_configs = [
-          {
-            targets = [
-              "tok-00.namely.icu:8080"
-              "la-00.video.namely.icu:8080"
-              "auth.xinyang.life:443"
-              "home.xinyang.life:8000"
-            ];
-          }
-        ];
-        relabel_configs = [
-          {
-            source_labels = [ "__address__" ];
-            target_label = "__param_target";
-          }
-          {
-            source_labels = [ "__param_target" ];
-            target_label = "instance";
-          }
-          {
-            target_label = "__address__";
-            replacement = "127.0.0.1:${toString config.services.prometheus.exporters.blackbox.port}";
-          }
-        ];
-      }
-      {
-        job_name = "blackbox_exporter";
-        static_configs = [
-          { targets = [ "127.0.0.1:${toString config.services.prometheus.exporters.blackbox.port}" ]; }
-        ];
-      }
-    ];
-
-    custom.prometheus.ruleModules = [
-      {
-        name = "probe_alerts";
-        rules = [
-          {
-            alert = "HighProbeLatency";
-            expr = "probe_duration_seconds > 0.5";
-            for = "2m";
-            labels = {
-              severity = "warning";
-            };
-            annotations = {
-              summary = "High request latency on {{ $labels.instance }}";
-              description = "95th percentile of request latency is above 0.5 seconds for the last 2 minutes.";
-            };
-          }
-        ];
-      }
-    ];
-  };
-}

modules/nixos/prometheus/caddy.nix

@@ -1,40 +0,0 @@
-{ config, lib, ... }:
-let
-  cfg = config.custom.prometheus;
-in
-{
-  config = lib.mkIf (cfg.enable && cfg.exporters.caddy.enable) {
-    services.caddy.globalConfig = lib.mkIf cfg.exporters.caddy.enable ''
-      servers {
-        metrics
-      }
-    '';
-
-    services.prometheus.scrapeConfigs = [
-      {
-        job_name = "caddy";
-        static_configs = [ { targets = [ "127.0.0.1:2019" ]; } ];
-      }
-    ];
-
-    custom.prometheus.ruleModules = [
-      {
-        name = "caddy_alerts";
-        rules = [
-          {
-            alert = "UpstreamHealthy";
-            expr = "caddy_reverse_proxy_upstreams_healthy != 1";
-            for = "5m";
-            labels = {
-              severity = "critical";
-            };
-            annotations = {
-              summary = "Upstream {{ $labels.unstream }} not healthy";
-            };
-          }
-        ];
-      }
-    ];
-  };
-
-}

modules/nixos/prometheus/default.nix

@@ -1,21 +1,16 @@
 {
   config,
-  pkgs,
   lib,
   ...
 }:
-
-with lib;
-
 let
+  inherit (lib)
+    mkEnableOption
+    mkOption
+    mkIf
+    types
+    ;
   cfg = config.custom.prometheus;
-  mkExporterOption =
-    enableOption:
-    (mkOption {
-      type = types.bool;
-      default = enableOption;
-      description = "Enable this exporter";
-    });
 
   mkRulesOption = mkOption {
     type = types.listOf (
@@ -30,38 +25,36 @@ let
 in
 {
   imports = [
-    ./blackbox.nix
+    ./exporters.nix
-    ./caddy.nix
+    ./grafana.nix
-    ./gotosocial.nix
-    ./immich.nix
-    ./miniflux.nix
-    ./ntfy-sh.nix
-    ./restic.nix
   ];
 
   options = {
+    custom.monitoring = {
+      grafana = {
+        enable = mkEnableOption "grafana with oauth only";
+      };
+    };
     custom.prometheus = {
       enable = mkEnableOption "Prometheus instance";
-      exporters = {
-        enable = mkOption {
-          type = types.bool;
-          default = false;
-          description = "Enable Prometheus exporter on every supported services";
-        };
-
-        restic.enable = mkExporterOption config.services.restic.server.enable;
-        blackbox.enable = mkExporterOption false;
-        caddy.enable = mkExporterOption config.services.caddy.enable;
-        gotosocial.enable = mkExporterOption config.services.gotosocial.enable;
-        immich.enable = mkExporterOption config.services.immich.enable;
-        miniflux.enable = mkExporterOption config.services.miniflux.enable;
-        ntfy-sh.enable = mkExporterOption config.services.ntfy-sh.enable;
-      };
-      grafana = {
-        enable = mkEnableOption "Grafana Cloud";
-        password_file = mkOption { type = types.path; };
-      };
       ruleModules = mkRulesOption;
+      exporters = {
+        enable = mkEnableOption "prometheus exporter on all supported and enable guarded services";
+        node = {
+          enable = mkEnableOption "node exporter";
+          listenAddress = mkOption {
+            type = types.str;
+            default = "${config.networking.hostName}.coho-tet.ts.net";
+          };
+        };
+        blackbox = {
+          enable = mkEnableOption "blackbox exporter";
+          listenAddress = mkOption {
+            type = types.str;
+            default = "${config.networking.hostName}.coho-tet.ts.net";
+          };
+        };
+      };
     };
   };
 
@@ -78,46 +71,18 @@ in
         reverse_proxy 127.0.0.1:${toString config.services.prometheus.port}
       '';
     };
-
     services.prometheus = mkIf cfg.enable {
       enable = true;
       port = 9091;
       globalConfig.external_labels = {
         hostname = config.networking.hostName;
       };
-      remoteWrite = mkIf cfg.grafana.enable [
+
-        {
-          name = "grafana";
-          url = "https://prometheus-prod-24-prod-eu-west-2.grafana.net/api/prom/push";
-          basic_auth = {
-            username = "1340065";
-            password_file = cfg.grafana.password_file;
-          };
-        }
-      ];
-      exporters = {
-        node = {
-          enable = true;
-          enabledCollectors = [
-            "loadavg"
-            "time"
-            "systemd"
-          ];
-          listenAddress = "127.0.0.1";
-          port = 9100;
-        };
-      };
       scrapeConfigs = [
         {
           job_name = "prometheus";
           static_configs = [ { targets = [ "localhost:${toString config.services.prometheus.port}" ]; } ];
         }
-        {
-          job_name = "node";
-          static_configs = [
-            { targets = [ "localhost:${toString config.services.prometheus.exporters.node.port}" ]; }
-          ];
-        }
       ];
 
       alertmanager = {
@@ -163,55 +128,8 @@ in
     };
     custom.prometheus.ruleModules = [
       {
-        name = "system_alerts";
+        name = "prometheus_alerts";
         rules = [
-          {
-            alert = "SystemdFailedUnits";
-            expr = "node_systemd_unit_state{state=\"failed\"} > 0";
-            for = "5m";
-            labels = {
-              severity = "critical";
-            };
-            annotations = {
-              summary = "Systemd has failed units on {{ $labels.instance }}";
-              description = "There are {{ $value }} failed units on {{ $labels.instance }}. Immediate attention required!";
-            };
-          }
-          {
-            alert = "HighLoadAverage";
-            expr = "node_load1 > 0.8 * count without (cpu) (node_cpu_seconds_total{mode=\"idle\"})";
-            for = "1m";
-            labels = {
-              severity = "warning";
-            };
-            annotations = {
-              summary = "High load average detected on {{ $labels.instance }}";
-              description = "The 1-minute load average ({{ $value }}) exceeds 80% the number of CPUs.";
-            };
-          }
-          {
-            alert = "HighTransmitTraffic";
-            expr = "rate(node_network_transmit_bytes_total{device!=\"lo\"}[5m]) > 100000000";
-            for = "1m";
-            labels = {
-              severity = "warning";
-            };
-            annotations = {
-              summary = "High network transmit traffic on {{ $labels.instance }} ({{ $labels.device }})";
-              description = "The network interface {{ $labels.device }} on {{ $labels.instance }} is transmitting data at a rate exceeding 100 MB/s for the last 1 minute.";
-            };
-          }
-          {
-            alert = "NetworkTrafficExceedLimit";
-            expr = ''increase(node_network_transmit_bytes_total{device!="lo",device!~"tailscale.*",device!~"wg.*",device!~"br.*"}[30d]) > 322122547200'';
-            for = "0m";
-            labels = {
-              severity = "critical";
-            };
-            annotations = {
-              summary = "Outbound network traffic exceed 300GB for last 30 day";
-            };
-          }
           {
             alert = "JobDown";
             expr = "up == 0";

modules/nixos/prometheus/exporters.nix

@@ -0,0 +1,65 @@
+{
+  config,
+  pkgs,
+  lib,
+  ...
+}:
+let
+  inherit (lib) mkIf;
+  cfg = config.custom.prometheus.exporters;
+in
+{
+  config = {
+    services.prometheus.exporters.node = mkIf cfg.node.enable {
+      enable = true;
+      enabledCollectors = [
+        "loadavg"
+        "time"
+        "systemd"
+      ];
+      listenAddress = cfg.node.listenAddress;
+      port = 9100;
+    };
+
+    services.prometheus.exporters.blackbox = mkIf cfg.blackbox.enable {
+      enable = true;
+      listenAddress = cfg.blackbox.listenAddress;
+      configFile = pkgs.writeText "blackbox.config.yaml" (
+        lib.generators.toYAML { } {
+          modules = {
+            tcp4_connect = {
+              prober = "tcp";
+              tcp = {
+                ip_protocol_fallback = false;
+                preferred_ip_protocol = "ip4";
+                tls = false;
+              };
+              timeout = "15s";
+            };
+          };
+        }
+      );
+    };
+
+    services.gotosocial.settings = {
+      metrics-enabled = true;
+    };
+
+    services.immich.environment = {
+      IMMICH_TELEMETRY_INCLUDE = "all";
+    };
+
+    services.restic.server.prometheus = true;
+    systemd.services.miniflux.environment.METRICS_COLLECTOR = "1";
+    services.ntfy-sh.settings.enable-metrics = true;
+
+    services.caddy.globalConfig = ''
+      servers {
+        metrics
+      }
+
+      admin ${config.networking.hostName}.coho-tet.ts.net:2019 {
+      }
+    '';
+  };
+}

modules/nixos/prometheus/gotosocial.nix

@@ -1,17 +0,0 @@
-{ config, lib, ... }:
-let
-  cfg = config.custom.prometheus;
-in
-{
-  config = lib.mkIf (cfg.enable && cfg.exporters.gotosocial.enable) {
-    services.gotosocial.settings = {
-      metrics-enabled = true;
-    };
-    services.prometheus.scrapeConfigs = [
-      {
-        job_name = "gotosocial";
-        static_configs = [ { targets = [ "localhost:8080" ]; } ];
-      }
-    ];
-  };
-}

modules/nixos/prometheus/grafana.nix

@@ -0,0 +1,43 @@
+{ config, lib, ... }:
+let
+  cfg = config.custom.monitoring.grafana;
+in
+{
+  config = lib.mkIf cfg.enable {
+    sops.templates."grafana.env".content = ''
+      GF_AUTH_GENERIC_OAUTH_CLIENT_SECRET=${config.sops.placeholder."grafana/oauth_secret"}
+    '';
+    services.grafana = {
+      enable = true;
+      settings = {
+        server = {
+          http_addr = "127.0.0.1";
+          http_port = 3003;
+          root_url = "https://grafana.xinyang.life";
+          domain = "grafana.xinyang.life";
+        };
+        "auth.generic_oauth" = {
+          enabled = true;
+          name = "Kanidm";
+          client_id = "grafana";
+          scopes = "openid,profile,email,groups";
+          auth_url = "https://auth.xinyang.life/ui/oauth2";
+          token_url = "https://auth.xinyang.life/oauth2/token";
+          api_url = "https://auth.xinyang.life/oauth2/openid/grafana/userinfo";
+          use_pkce = true;
+          use_refresh_token = true;
+          allow_sign_up = true;
+          login_attribute_path = "preferred_username";
+          groups_attribute_path = "groups";
+          role_attribute_path = "contains(grafana_role[*], 'GrafanaAdmin') && 'GrafanaAdmin' || contains(grafana_role[*], 'Admin') && 'Admin' || contains(grafana_role[*], 'Editor') && 'Editor' || 'Viewer'";
+          allow_assign_grafana_admin = true;
+          auto_login = true;
+        };
+        "auth" = {
+          disable_login_form = true;
+        };
+      };
+    };
+    systemd.services.grafana.serviceConfig.EnvironmentFile = config.sops.templates."grafana.env".path;
+  };
+}

modules/nixos/prometheus/immich.nix

@@ -1,25 +0,0 @@
-{ config, lib, ... }:
-let
-  cfg = config.custom.prometheus;
-  immichEnv = config.services.immich.environment;
-  metricPort =
-    if builtins.hasAttr "IMMICH_API_METRICS_PORT" immichEnv then
-      immichEnv.IMMICH_API_METRICS_PORT
-    else
-      8081;
-in
-{
-  config = lib.mkIf (cfg.enable && cfg.exporters.immich.enable) {
-    services.immich.environment = {
-      IMMICH_METRICS = "true";
-    };
-
-    services.prometheus.scrapeConfigs = [
-      {
-        job_name = "immich";
-        static_configs = [ { targets = [ "127.0.0.1:${toString metricPort}" ]; } ];
-      }
-    ];
-  };
-
-}

modules/nixos/prometheus/miniflux.nix

@@ -1,15 +0,0 @@
-{ config, lib, ... }:
-let
-  cfg = config.custom.prometheus;
-in
-{
-  config = lib.mkIf (cfg.enable && cfg.exporters.miniflux.enable) {
-    systemd.services.miniflux.environment.METRICS_COLLECTOR = "1";
-    services.prometheus.scrapeConfigs = [
-      {
-        job_name = "miniflux";
-        static_configs = [ { targets = [ config.systemd.services.miniflux.environment.LISTEN_ADDR ]; } ];
-      }
-    ];
-  };
-}

modules/nixos/prometheus/ntfy-sh.nix

@@ -1,15 +0,0 @@
-{ config, lib, ... }:
-let
-  cfg = config.custom.prometheus;
-in
-{
-  config = lib.mkIf (cfg.enable && cfg.exporters.ntfy-sh.enable) {
-    services.ntfy-sh.settings.enable-metrics = true;
-    services.prometheus.scrapeConfigs = [
-      {
-        job_name = "ntfy-sh";
-        static_configs = [ { targets = [ "ntfy.xinyang.life" ]; } ];
-      }
-    ];
-  };
-}

modules/nixos/prometheus/restic.nix

@@ -3,17 +3,26 @@ let
   cfg = config.custom.prometheus;
 in
 {
-  config = lib.mkIf (cfg.enable && cfg.exporters.restic.enable) {
+  config = {
     services.restic.server.prometheus = true;
 
-    services.prometheus.scrapeConfigs = [
+    custom.prometheus.templates.scrape.mkResticScrapes =
-      (lib.mkIf cfg.exporters.restic.enable {
+      {
-        job_name = "restic";
+        address,
-        static_configs = [ { targets = [ config.services.restic.server.listenAddress ]; } ];
+        port ? null,
-      })
+        ...
-    ];
+      }:
+      let
+        portStr = if port then ":${toString port}" else "";
+      in
+      [
+        (lib.mkIf cfg.exporters.restic.enable {
+          job_name = "restic";
+          static_configs = [ { targets = [ "${address}${portStr}" ]; } ];
+        })
+      ];
 
-    custom.prometheus.ruleModules = [
+    custom.prometheus.templates.rules.mkResticRules = [
       {
         name = "restic_alerts";
         rules = [

overlays/my-lib/default.nix

@@ -0,0 +1,3 @@
+{
+}
+// (import ./prometheus.nix)

overlays/my-lib/prometheus.nix

@@ -0,0 +1,199 @@
+let
+  mkFunction = f: (targets: (map f targets));
+  mkPort = port: if isNull port then "" else ":${toString port}";
+in
+{
+  mkScrapes = mkFunction (
+    {
+      name,
+      address,
+      port ? 443,
+      scheme ? "https",
+      ...
+    }:
+    {
+      job_name = "${name}(${address})";
+      scheme = scheme;
+      static_configs = [ { targets = [ "${address}${mkPort port}" ]; } ];
+    }
+  );
+
+  mkCaddyScrapes = mkFunction (
+    {
+      address,
+      port ? 2019,
+      ...
+    }:
+    {
+      job_name = "caddy_${address}";
+      static_configs = [ { targets = [ "${address}${mkPort port}" ]; } ];
+    }
+  );
+
+  mkCaddyRules = mkFunction (
+    {
+      host ? "",
+      name ? "caddy_alerts_${host}",
+    }:
+    {
+      inherit name;
+      rules = [
+        {
+          alert = "UpstreamHealthy";
+          expr = "caddy_reverse_proxy_upstreams_healthy != 1";
+          for = "5m";
+          labels = {
+            severity = "critical";
+          };
+          annotations = {
+            summary = "Upstream {{ $labels.unstream }} not healthy";
+          };
+        }
+      ];
+    }
+  );
+
+  mkNodeScrapes = mkFunction (
+    {
+      address,
+      port ? 9100,
+      ...
+    }:
+    {
+      job_name = "node_${address}";
+      static_configs = [ { targets = [ "${address}${mkPort port}" ]; } ];
+    }
+  );
+
+  mkNodeRules = mkFunction (
+    {
+      host ? "",
+      name ? "system_alerts_${host}",
+      ...
+    }:
+    {
+      inherit name;
+      rules = [
+        {
+          alert = "SystemdFailedUnits";
+          expr = "node_systemd_unit_state{state=\"failed\"} > 0";
+          for = "5m";
+          labels = {
+            severity = "critical";
+          };
+          annotations = {
+            summary = "Systemd has failed units on {{ $labels.instance }}";
+            description = "There are {{ $value }} failed units on {{ $labels.instance }}. Immediate attention required!";
+          };
+        }
+        {
+          alert = "HighLoadAverage";
+          expr = "node_load1 > 0.8 * count without (cpu) (node_cpu_seconds_total{mode=\"idle\"})";
+          for = "1m";
+          labels = {
+            severity = "warning";
+          };
+          annotations = {
+            summary = "High load average detected on {{ $labels.instance }}";
+            description = "The 1-minute load average ({{ $value }}) exceeds 80% the number of CPUs.";
+          };
+        }
+        {
+          alert = "HighTransmitTraffic";
+          expr = "rate(node_network_transmit_bytes_total{device!=\"lo\"}[5m]) > 100000000";
+          for = "1m";
+          labels = {
+            severity = "warning";
+          };
+          annotations = {
+            summary = "High network transmit traffic on {{ $labels.instance }} ({{ $labels.device }})";
+            description = "The network interface {{ $labels.device }} on {{ $labels.instance }} is transmitting data at a rate exceeding 100 MB/s for the last 1 minute.";
+          };
+        }
+        {
+          alert = "NetworkTrafficExceedLimit";
+          expr = ''increase(node_network_transmit_bytes_total{device!="lo",device!~"tailscale.*",device!~"wg.*",device!~"br.*"}[30d]) > 322122547200'';
+          for = "0m";
+          labels = {
+            severity = "critical";
+          };
+          annotations = {
+            summary = "Outbound network traffic exceed 300GB for last 30 day";
+          };
+        }
+      ];
+    }
+  );
+
+  mkBlackboxScrapes = mkFunction (
+    {
+      hostAddress,
+      hostPort ? 9115,
+      targetAddresses,
+      ...
+    }:
+    {
+      job_name = "blackbox(${hostAddress})";
+      scrape_interval = "1m";
+      metrics_path = "/probe";
+      params = {
+        module = [ "tcp4_connect" ];
+      };
+      static_configs = [
+        {
+          targets = targetAddresses;
+        }
+      ];
+      relabel_configs = [
+        {
+          source_labels = [ "__address__" ];
+          target_label = "__param_target";
+        }
+        {
+          source_labels = [ "__param_target" ];
+          target_label = "instance";
+        }
+        {
+          target_label = "__address__";
+          replacement = "${hostAddress}${mkPort hostPort}";
+        }
+      ];
+    }
+  );
+
+  mkBlackboxRules = mkFunction (
+    {
+      host ? "",
+      name ? "probe_alerts_${host}",
+    }:
+    {
+      inherit name;
+      rules = [
+        {
+          alert = "HighProbeLatency";
+          expr = "probe_duration_seconds > 0.5";
+          for = "3m";
+          labels = {
+            severity = "warning";
+          };
+          annotations = {
+            summary = "High request latency on {{ $labels.instance }}";
+            description = "Request latency is above 0.5 seconds for the last 3 minutes.";
+          };
+        }
+        {
+          alert = "VeryHighProbeLatency";
+          expr = "probe_duration_seconds > 1";
+          for = "3m";
+          labels = {
+            severity = "critical";
+          };
+          annotations = {
+            summary = "High request latency on {{ $labels.instance }}";
+            description = "Request latency is above 0.5 seconds for the last 3 minutes.";
+          };
+        }
+      ];
+    }
+  );
+}