cluster/services/ipfs/monitoring.nix

@@ -9,7 +9,7 @@ in
     environment = {
       OTEL_TRACES_EXPORTER = "otlp";
       OTEL_EXPORTER_OTLP_PROTOCOL = "grpc";
-      OTEL_EXPORTER_OTLP_ENDPOINT = cluster.config.ways.ingest-traces-otlp.url;
+      OTEL_EXPORTER_OTLP_ENDPOINT = cluster.config.links.tempo-otlp-grpc.url;
       OTEL_TRACES_SAMPLER = "parentbased_traceidratio";
       OTEL_TRACES_SAMPLER_ARG = "0.50";
     };

cluster/services/monitoring/default.nix

@@ -18,6 +18,26 @@ in
       protocol = "http";
       ipv4 = meshIpFor "server";
     };
+    tempo = {
+      protocol = "http";
+      ipv4 = meshIpFor "server";
+    };
+    tempo-grpc = {
+      protocol = "http";
+      ipv4 = "127.0.0.1";
+    };
+    tempo-otlp-http = {
+      protocol = "http";
+      ipv4 = meshIpFor "server";
+    };
+    tempo-otlp-grpc = {
+      protocol = "http";
+      ipv4 = meshIpFor "server";
+    };
+    tempo-zipkin-http = {
+      protocol = "http";
+      ipv4 = meshIpFor "server";
+    };
   };
   hostLinks = lib.genAttrs config.services.monitoring.nodes.grafana (name: {
     grafana = {
@@ -31,7 +51,6 @@ in
       blackbox = [ "checkmate" "grail" "prophet" ];
       grafana = [ "VEGAS" "prophet" ];
       logging = [ "VEGAS" "grail" ];
-      tracing = [ "VEGAS" "grail" ];
       server = [ "VEGAS" ];
     };
     nixos = {
@@ -42,20 +61,12 @@ in
         ./provisioning/dashboards.nix
       ];
       logging = ./logging.nix;
-      tracing = ./tracing.nix;
       server = [
         ./server.nix
+        ./tracing.nix
       ];
     };
-    meshLinks = {
-      logging.loki.link.protocol = "http";
-      tracing = {
-        tempo.link.protocol = "http";
-        tempo-otlp-http.link.protocol = "http";
-        tempo-otlp-grpc.link.protocol = "grpc";
-        tempo-zipkin-http.link.protocol = "http";
-      };
-    };
+    meshLinks.logging.loki.link.protocol = "http";
   };
 
   garage = config.lib.forService "monitoring" {
@@ -68,51 +79,36 @@ in
         nodes = config.services.monitoring.nodes.logging;
         format = "envFile";
       };
-      tempo-ingest.locksmith = {
-        nodes = config.services.monitoring.nodes.tracing;
-        format = "envFile";
-      };
-      tempo-query.locksmith = {
-        nodes = config.services.monitoring.nodes.tracing;
-        format = "envFile";
-      };
+      tempo = { };
     };
     buckets = {
       loki-chunks.allow = {
         loki-ingest = [ "read" "write" ];
         loki-query = [ "read" ];
       };
-      tempo-chunks.allow = {
-        tempo-ingest = [ "read" "write" ];
-        tempo-query = [ "read" ];
-      };
+      tempo-chunks.allow.tempo = [ "read" "write" ];
     };
   };
 
-  ways = let
-    query = consulService: {
-      inherit consulService;
+  ways = config.lib.forService "monitoring" {
+    monitoring = {
+      consulService = "grafana";
+      extras.locations."/".proxyWebsockets = true;
+    };
+    monitoring-logs = {
       internal = true;
+      consulService = "loki";
       extras.extraConfig = ''
         proxy_read_timeout 3600s;
       '';
     };
-    ingest = consulService: {
-      inherit consulService;
+    ingest-logs = {
       internal = true;
+      consulService = "loki";
       extras.extraConfig = ''
         client_max_body_size 4G;
         proxy_read_timeout 3600s;
       '';
     };
-  in config.lib.forService "monitoring" {
-    monitoring = {
-      consulService = "grafana";
-      extras.locations."/".proxyWebsockets = true;
-    };
-    monitoring-logs = query "loki";
-    monitoring-traces = query "tempo";
-    ingest-logs = ingest "loki";
-    ingest-traces-otlp = ingest "tempo-ingest-otlp-grpc" // { grpc = true; };
   };
 }

cluster/services/monitoring/grafana-ha.nix

@@ -73,16 +73,6 @@ in
           inherit (cluster.config.ways.monitoring-logs) url;
           type = "loki";
         }
-        {
-          name = "Tempo";
-          uid = "P214B5B846CF3925F";
-          inherit (cluster.config.ways.monitoring-traces) url;
-          type = "tempo";
-          jsonData = {
-            serviceMap.datasourceUid = "PBFA97CFB590B2093";
-            nodeGraph.enabled = true;
-          };
-        }
       ];
     };
   };

cluster/services/monitoring/secrets/tempo-secrets.age

@@ -0,0 +1,16 @@
+age-encryption.org/v1
+-> ssh-ed25519 NO562A KhCGp7PAMGrEdzRxBrsW4tRk30JwpI+4lPzrRCUhSw4
+8s7WqA5c3zS1euN5R+jfFNBdvr8OQW8P4NFeqtNsIKo
+-> ssh-ed25519 5/zT0w 79hJQ2H76EZTW7YcQFCtKaS5Kbssx4Z8dPFjIVzRgFk
+A1fDJbUnyIRy+kWa3PhJNj/SdRPlcEy6FYsAfnuZ2AQ
+-> ssh-ed25519 d3WGuA aylkdL1KliM1NfrYDGlG8X6YjXvVUCU4sV90I+a840U
+6sXdqIPjtoNSylZRh1DCghHOwDo+fC7WB4QWQoWmG48
+-> //gd+2-grease baUWA$3 z-qs3W O/2.1W
+Sfq3+rkMJhpUTTmcos5TaaUtX2Ip9pciHAZLiWPix+C9N7ccac/1W5RNedMJCLsq
+MQ+xKzexf8+hgNVhKOksvbKBBROXqk1bUOKk8w3OgFPmmByzmCBUwkdkeu5DFTYR
+rg
+--- kUl1uIPRkM5y7C68kdN22pMKXP7gazyha4PE+ap0Jqw
+w>×Àè¥
+<15>CÈ,\‰ßœI¯ˆúHxG@^Çá“PåÃþÙÏlw6
µŽ{þ’rb‘é5æ†T>Êñ
+ÚWܤX4Kp(ß?9ˆß­^^oP3f </v3N$ê¤sÓbŽ¾>O™÷œ+òN0άïµàDtêŽ5Vº#è ¶³‘Uã îŸ#y|›@ŒGzSi»­ô*·HùüŽ]
+ꎀ5›

cluster/services/monitoring/tracing.nix

@@ -1,16 +1,14 @@
 { cluster, config, pkgs, ... }:
 
 let
-  inherit (cluster.config.links) prometheus-ingest;
-  inherit (config.links) tempo-grpc;
-  links = cluster.config.hostLinks.${config.networking.hostName};
+  inherit (cluster.config) links;
   dataDir = "/srv/storage/private/tempo";
   tempoConfig = {
     server = {
       http_listen_address = links.tempo.ipv4;
       http_listen_port = links.tempo.port;
-      grpc_listen_address = tempo-grpc.ipv4;
-      grpc_listen_port = tempo-grpc.port;
+      grpc_listen_address = links.tempo-grpc.ipv4;
+      grpc_listen_port = links.tempo-grpc.port;
     };
     distributor.receivers = {
       otlp = {
@@ -21,7 +19,7 @@ let
       };
       zipkin.endpoint = links.tempo-zipkin-http.tuple;
     };
-    querier.frontend_worker.frontend_address = tempo-grpc.tuple;
+    querier.frontend_worker.frontend_address = links.tempo-grpc.tuple;
     ingester = {
       trace_idle_period = "30s";
       max_block_bytes = 1000000;
@@ -58,7 +56,7 @@ let
         path = "${dataDir}/generator/wal";
         remote_write = [
           {
-            url = "${prometheus-ingest.url}/api/v1/write";
+            url = "${links.prometheus-ingest.url}/api/v1/write";
             send_exemplars = true;
           }
         ];
@@ -70,11 +68,7 @@ let
     ];
   };
 in {
-  links.tempo-grpc.protocol = "http";
-
-  services.locksmith.waitForSecrets.tempo = [
-    "garage-tempo-ingest"
-  ];
+  age.secrets.tempoSecrets.file = ./secrets/tempo-secrets.age;
 
   users.users.tempo = {
     isSystemUser = true;
@@ -87,53 +81,24 @@ in {
 
   systemd.services.tempo = {
     wantedBy = [ "multi-user.target" ];
-    distributed = {
-      enable = true;
-      registerServices = [
-        "tempo"
-        "tempo-ingest-otlp-grpc"
-      ];
-    };
     serviceConfig = {
       User = "tempo";
       Group = "tempo";
       ExecStart = "${pkgs.tempo}/bin/tempo -config.file=${pkgs.writeText "tempo.yaml" (builtins.toJSON tempoConfig)}";
       PrivateTmp = true;
-      EnvironmentFile = "/run/locksmith/garage-tempo-ingest";
+      EnvironmentFile = config.age.secrets.tempoSecrets.path;
     };
   };
-
-  consul.services = {
-    tempo = {
-      mode = "manual";
-      definition = {
-        name = "tempo";
-        address = links.tempo.ipv4;
-        inherit (links.tempo) port;
-        checks = [
-          {
-            name = "Tempo";
-            id = "service:tempo:backend";
-            interval = "5s";
-            http = "${links.tempo.url}/ready";
-          }
-        ];
+  services.grafana.provision.datasources.settings.datasources = [
+    {
+      name = "Tempo";
+      uid = "P214B5B846CF3925F";
+      inherit (links.tempo) url;
+      type = "tempo";
+      jsonData = {
+        serviceMap.datasourceUid = "PBFA97CFB590B2093"; # prometheus
+        nodeGraph.enabled = true;
       };
-    };
-    tempo-ingest-otlp-grpc = {
-      mode = "manual";
-      definition = {
-        name = "tempo-ingest-otlp-grpc";
-        address = links.tempo-otlp-grpc.ipv4;
-        inherit (links.tempo-otlp-grpc) port;
-        checks = [
-          {
-            name = "Tempo Service Status";
-            id = "service:tempo-ingest-otlp-grpc:tempo";
-            alias_service = "tempo";
-          }
-        ];
-      };
-    };
-  };
+    }
+  ];
 }

cluster/services/sso/host.nix

@@ -54,7 +54,7 @@ in
       "-Dotel.traces.exporter=otlp"
     ];
     OTEL_EXPORTER_OTLP_PROTOCOL = "grpc";
-    OTEL_EXPORTER_OTLP_ENDPOINT = cluster.config.ways.ingest-traces-otlp.url;
+    OTEL_EXPORTER_OTLP_ENDPOINT = cluster.config.links.tempo-otlp-grpc.url;
     OTEL_TRACES_SAMPLER = "parentbased_traceidratio";
     OTEL_TRACES_SAMPLER_ARG = "0.50";
   };

cluster/services/ways/host.nix

@@ -25,14 +25,7 @@ in
             ];
             locations = lib.mkMerge [
               {
-                "/" = if cfg.grpc then {
-                  extraConfig = ''
-                    set $nix_proxy_grpc_target ${cfg.target};
-                    grpc_pass $nix_proxy_grpc_target;
-                  '';
-                } else {
-                  proxyPass = cfg.target;
-                };
+                "/".proxyPass = cfg.target;
                 "${cfg.healthCheckPath}".extraConfig = "access_log off;";
               }
               {

cluster/services/ways/options/way.nix

@@ -35,12 +35,6 @@ with lib;
       };
     };
 
-    grpc = mkOption {
-      description = "Whether this endpoint is a gRPC service.";
-      type = types.bool;
-      default = false;
-    };
-
     target = mkOption {
       type = types.str;
     };
@@ -107,7 +101,7 @@ with lib;
     (lib.mkIf options.consulService.isDefined {
       useConsul = true;
       nginxUpstreamName = "ways_upstream_${builtins.hashString "md5" options.consulService.value}";
-      target = "${if config.grpc then "grpc" else "http"}://${options.nginxUpstreamName.value}";
+      target = "http://${options.nginxUpstreamName.value}";
     })
     (lib.mkIf options.bucket.isDefined {
       consulService = "garage-web";

modules/consul-distributed-services/default.nix

@@ -26,13 +26,11 @@ in
 
       cfg = v.distributed;
 
-      svcs = map (x: config.consul.services.${x}) cfg.registerServices;
+      svc = config.consul.services.${cfg.registerService};
 
       runWithRegistration = pkgs.writeShellScript "run-with-registration" ''
-        trap '${lib.concatStringsSep ";" (map (svc: svc.commands.deregister) svcs)}' EXIT
-        ${lib.concatStringsSep "\n" (
-          map (svc: svc.commands.register) svcs
-        )}
+        trap '${svc.commands.deregister}' EXIT
+        ${svc.commands.register}
         ''${@}
       '';
 
@@ -51,10 +49,10 @@ in
       [Service]
       ExecStartPre=${waitForConsul} 'services/${n}%i'
       ExecStart=
-      ExecStart=${consul}/bin/consul lock --name=${n} --n=${toString cfg.replicas} --shell=false --child-exit-code 'services/${n}%i' ${optionalString (cfg.registerServices != []) runWithRegistration} ${ExecStart}
+      ExecStart=${consul}/bin/consul lock --name=${n} --n=${toString cfg.replicas} --shell=false --child-exit-code 'services/${n}%i' ${optionalString (cfg.registerService != null) runWithRegistration} ${ExecStart}
       Environment="CONSUL_HTTP_ADDR=${consulHttpAddr}"
       ${optionalString (v.serviceConfig ? RestrictAddressFamilies) "RestrictAddressFamilies=AF_NETLINK"}
-      ${optionalString (cfg.registerServices != []) (lib.concatStringsSep "\n" (map (svc: "ExecStopPost=${svc.commands.deregister}") svcs))}
+      ${optionalString (cfg.registerService != null) "ExecStopPost=${svc.commands.deregister}"}
     ''))
   ];
 }

modules/systemd-extras/distributed.nix

@@ -17,11 +17,6 @@ with lib;
           type = with types; nullOr str;
           default = null;
         };
-        registerServices = mkOption {
-          description = "Consul services to register when this service gets started.";
-          type = with types; listOf str;
-          default = if config.distributed.registerService == null then [ ] else [ config.distributed.registerService ];
-        };
       };
     }));
   };