Monitoring

jobs/ingress/traefik.hcl

@@ -44,6 +44,10 @@ job "traefik" {
           "/storage/nomad/traefik/access.log:/access.log",
         ]
       }
+      resources {
+        cpu    = 500
+        memory = 1024
+      }
 
       template {
         destination = "local/.env"
@@ -137,6 +141,12 @@ EOF
 
 [tracing]
 
+[metrics]
+  [metrics.prometheus]
+    addServicesLabels = true
+    addRoutersLabels = true
+    addEntryPointsLabels = true
+
 [accessLog]
   filePath = "/access.log"
 EOF
@@ -221,6 +231,7 @@ EOF
       certResolver = "lets-encrypt"
 
 {{ end -}}
+      url = "http://127.0.0.1"  # Dummy service - not used
 EOF
       }
     }

jobs/monitoring/grafana.hcl

@@ -0,0 +1,110 @@
+job "grafana" {
+  datacenters = ["aperture"]
+
+  type = "service"
+
+  group "monitoring" {
+    network {
+      port "http" {
+        to = 3000
+      }
+      port "db" {
+        to = 5432
+      }
+    }
+
+    service {
+      name = "grafana"
+      port = "http"
+
+      check {
+        type     = "http"
+        path     = "/"
+        interval = "10s"
+        timeout  = "2s"
+      }
+
+      tags = [
+        "traefik.enable=true",
+        "traefik.http.routers.grafana.entrypoints=web,websecure",
+        "traefik.http.routers.grafana.rule=Host(`grafana.redbrick.dcu.ie`)",
+        "traefik.http.routers.grafana.tls=true",
+        "traefik.http.routers.grafana.tls.certresolver=lets-encrypt",
+      ]
+    }
+
+    task "grafana" {
+      driver = "docker"
+      user   = "1001:1001"
+
+      env {
+        GF_AUTH_BASIC_ENABLED = "true"
+        GF_INSTALL_PLUGINS    = "grafana-piechart-panel"
+        GF_SERVER_ROOT_URL    = "https://grafana.redbrick.dcu.ie"
+      }
+
+      config {
+        image = "grafana/grafana"
+        ports = ["http"]
+
+        volumes = [
+          "/storage/nomad/${NOMAD_JOB_NAME}/${NOMAD_TASK_NAME}:/var/lib/grafana",
+          "local/datasources.yml:/etc/grafana/provisioning/datasources/datasources.yml"
+        ]
+      }
+
+      template {
+        data        = <<EOH
+GF_DATABASE_TYPE=postgres
+GF_DATABASE_HOST={{ env "NOMAD_ADDR_db" }}
+GF_DATABASE_NAME={{ key "grafana/db/name" }}
+GF_DATABASE_USER={{ key "grafana/db/user" }}
+GF_DATABASE_PASSWORD={{ key "grafana/db/password" }}
+GF_FEATURE_TOGGLES_ENABLE=publicDashboards
+GF_LOG_LEVEL=debug
+EOH
+        destination = "local/.env"
+        env         = true
+      }
+      template {
+        data        = <<EOH
+apiVersion: 1
+
+datasources:
+  - name: Prometheus
+    type: prometheus
+    access: proxy
+    {{- range service "prometheus" }}
+    url: http://prometheus.service.consul:{{ .Port }}{{ end }}
+    isDefault: true
+    editable: false
+EOH
+        destination = "local/datasources.yml"
+      }
+    }
+    task "db" {
+      driver = "docker"
+
+      config {
+        image = "postgres:17-alpine"
+        ports = ["db"]
+
+        volumes = [
+          "/storage/nomad/${NOMAD_JOB_NAME}/${NOMAD_TASK_NAME}:/var/lib/postgresql/data",
+        ]
+      }
+
+      template {
+        data        = <<EOH
+POSTGRES_PASSWORD={{ key "grafana/db/password" }}
+POSTGRES_USER={{ key "grafana/db/user" }}
+POSTGRES_NAME={{ key "grafana/db/name" }}
+EOH
+        destination = "local/db.env"
+        env         = true
+      }
+    }
+  }
+}
+
+

jobs/monitoring/prometheus.hcl

@@ -0,0 +1,74 @@
+job "prometheus" {
+  datacenters = ["aperture"]
+  type = "service"
+
+  group "prometheus" {
+    count = 1
+
+    network {
+      port "http" {
+        to = 9090
+      }
+    }
+
+    task "prometheus" {
+      driver = "docker"
+
+      service {
+        name = "prometheus"
+        port = "http"
+      }
+
+      config {
+        image = "prom/prometheus:latest"
+        ports = ["http"]
+
+        volumes = [
+          "local/prometheus.yml:/etc/prometheus/prometheus.yml"
+        ]
+      }
+
+      template {
+        destination = "local/prometheus.yml"
+        data = <<EOF
+global:
+  scrape_interval: 10s
+  evaluation_interval: 10s
+
+scrape_configs:
+- job_name: 'nomad_metrics'
+  consul_sd_configs:
+    - server: 'consul.service.consul:8500'
+      services: ['nomad-client', 'nomad'] # This allows for Client (Workload) and Server (Orchastration) metrics
+      tags: ['http']
+
+  metrics_path: /v1/metrics
+  params:
+    format: ['prometheus']
+
+- job_name: 'container-metrics'
+  consul_sd_configs:
+  - server: 'consul.service.consul:8500'
+    tags: ['prometheus.enable=true']
+  relabel_configs:
+    - source_labels: ['__meta_consul_service']
+      target_label: 'job'
+      replacement: 'consul-service'
+    - source_labels: ['__meta_consul_tags']
+      regex: '.*prometheus.path=([^,]+).*'  # Extract path from tag
+      target_label: '__metrics_path__'
+
+  metrics_path: /v1/metrics
+  params:
+    format: ['prometheus']
+EOF
+      }
+
+      resources {
+        cpu    = 500
+        memory = 512
+      }
+    }
+  }
+}
+