massicot,calcite: enable blackbox exporter and probes

machines/calcite/configuration.nix

@@ -265,6 +265,11 @@
   custom.forgejo-actions-runner.enable = true;
   custom.forgejo-actions-runner.tokenFile = config.sops.secrets.gitea_env.path;
 
+  custom.prometheus = {
+    enable = true;
+    exporters.blackbox.enable = true;
+  };
+
   # MTP support
   services.gvfs.enable = true;
 

modules/nixos/prometheus/blackbox.nix

@@ -0,0 +1,83 @@
+{ 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

@@ -3,7 +3,7 @@ let
   cfg = config.custom.prometheus;
 in
 {
-  config = lib.mkIf cfg.enable {
+  config = lib.mkIf (cfg.enable && cfg.exporters.caddy.enable) {
     services.caddy.globalConfig = lib.mkIf cfg.exporters.caddy.enable ''
       servers {
         metrics
@@ -11,16 +11,16 @@ in
     '';
 
     services.prometheus.scrapeConfigs = [
-      (lib.mkIf cfg.exporters.caddy.enable {
+      {
         job_name = "caddy";
         static_configs = [
           { targets = [ "127.0.0.1:2019" ]; }
         ];
-      })
+      }
     ];
 
     custom.prometheus.ruleModules = [
-      (lib.mkIf cfg.exporters.caddy.enable {
+      {
         name = "caddy_alerts";
         rules = [
           {
@@ -38,7 +38,7 @@ in
             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/default.nix

@@ -4,7 +4,6 @@ with lib;
 
 let
   cfg = config.custom.prometheus;
-  exporterCfg = config.custom.prometheus.exporters;
   mkExporterOption = enableOption: (mkOption {
     type = types.bool;
     default = enableOption;
@@ -26,6 +25,7 @@ let
 in
 {
   imports = [
+    ./blackbox.nix
     ./caddy.nix
     ./gotosocial.nix
     ./ntfy-sh.nix
@@ -43,6 +43,7 @@ in
         };
 
         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;
         ntfy-sh.enable = mkExporterOption config.services.gotosocial.enable;
@@ -187,6 +188,13 @@ in
               labels = { severity = "critical"; };
               annotations = { summary = "Outbound network traffic exceed 300GB for last 30 day"; };
             }
+            {
+              alert = "JobDown";
+              expr = "up == 0";
+              for = "1m";
+              labels = { severity = "critical"; };
+              annotations = { summary = "Job {{ $labels.job }} down for 1m."; };
+            }
           ];
         }
       ];

modules/nixos/prometheus/gotosocial.nix

@@ -3,8 +3,8 @@ let
   cfg = config.custom.prometheus;
 in
 {
-  config = lib.mkIf cfg.exporters.gotosocial.enable {
-    services.gotosocial.settings = lib.mkIf cfg.exporters.gotosocial.enable {
+  config = lib.mkIf (cfg.enable && cfg.exporters.gotosocial.enable) {
+    services.gotosocial.settings = {
       metrics-enabled = true;
     };
     services.prometheus.scrapeConfigs = [

modules/nixos/prometheus/ntfy-sh.nix

@@ -3,15 +3,15 @@ let
   cfg = config.custom.prometheus;
 in
 {
-  config = lib.mkIf cfg.enable {
+  config = lib.mkIf (cfg.enable && cfg.exporters.ntfy-sh.enable) {
     services.ntfy-sh.settings.enable-metrics = true;
     services.prometheus.scrapeConfigs = [
-      (lib.mkIf cfg.exporters.ntfy-sh.enable {
+      {
         job_name = "ntfy-sh";
         static_configs = [
           { targets = [ "ntfy.xinyang.life" ]; }
         ];
-      })
+      }
     ];
   };
 }

modules/nixos/prometheus/restic.nix

@@ -3,7 +3,7 @@ let
   cfg = config.custom.prometheus;
 in
 {
-  config = lib.mkIf cfg.enable {
+  config = lib.mkIf (cfg.enable && cfg.exporters.restic.enable) {
     services.restic.server.prometheus = true;
 
     services.prometheus.scrapeConfigs = [
@@ -16,7 +16,7 @@ in
     ];
 
     custom.prometheus.ruleModules = [
-      (lib.mkIf cfg.exporters.restic.enable {
+      {
         name = "restic_alerts";
         rules = [
           {
@@ -34,7 +34,7 @@ in
             annotations = { summary = "Restic {{ $labels.client_hostname }} / {{ $labels.client_username }} backup is outdated"; description = "Restic backup is outdated\\n  VALUE = {{ $value }}\\n  LABELS = {{ $labels }}"; };
           }
         ];
-      })
+      }
     ];
   };