grafana: add dynamic prometheus datasource

jobs/games/minecraft/esports.hcl

@@ -1,64 +0,0 @@
-job "esports-minecraft" {
-  datacenters = ["aperture"]
-  type        = "service"
-
-  group "esports-mc" {
-    count = 1
-
-    network {
-      port "mc" {
-        to = 25565
-      }
-      port "rcon" {
-        to = 25575
-      }
-    }
-
-    service {
-      name = "esports-mc"
-      port = "mc"
-    }
-
-    service {
-      name = "esports-mc-rcon"
-      port = "rcon"
-    }
-
-    task "esports-minecraft" {
-      driver = "docker"
-      config {
-        image = "itzg/minecraft-server"
-        ports = ["mc", "rcon"]
-        volumes = [
-          "/storage/nomad/${NOMAD_TASK_NAME}:/data"
-        ]
-      }
-
-      resources {
-        cpu    = 5000  # 5000 MHz
-        memory = 20480 # 20 GB
-      }
-
-      template {
-        data        = <<EOF
-EULA            = "TRUE"
-TYPE            = "PAPER"
-VERSION         = "1.21.4"
-ICON            = "https://liquipedia.net/commons/images/thumb/5/53/DCU_Esports_allmode.png/37px-DCU_Esports_allmode.png"
-USE_AIKAR_FLAGS = true
-MAX_MEMORY      = 18G
-MOTD            = "Powered by Redbrick"
-MAX_PLAYERS     = "32"
-VIEW_DISTANCE   = "32"
-ENABLE_RCON     = true
-RCON_PASSWORD   = {{ key "games/mc/esports-mc/rcon/password" }}
-# Auto-download plugins
-SPIGET_RESOURCES=83581,62325,118271,28140,102931 # RHLeafDecay, GSit, GravesX, Luckperms, NoChatReport
-MODRINTH_PROJECTS=datapack:no-enderman-grief,thizzyz-tree-feller,imageframe,bmarker,datapack:players-drop-heads,viaversion,viabackwards
-EOF
-        destination = "local/.env"
-        env         = true
-      }
-    }
-  }
-}

jobs/ingress/traefik.hcl

@@ -51,7 +51,7 @@ job "traefik" {
 
     service {
       name     = "traefik-http"
-      provider = "nomad"
+      provider = "consul"
       port     = "https"
     }
 
@@ -67,6 +67,10 @@ job "traefik" {
           "/storage/nomad/traefik/access.log:/access.log",
         ]
       }
+      resources {
+        cpu    = 500
+        memory = 1024
+      }
 
       template {
         data        = <<EOF
@@ -157,7 +161,11 @@ job "traefik" {
   storage = "acme.json"
   [certificatesResolvers.lets-encrypt.acme.tlsChallenge]
 
-[tracing]
+[metrics]
+  [metrics.prometheus]
+    addServicesLabels = true
+    addRoutersLabels = true
+    addEntryPointsLabels = true
 
 [accessLog]
   filePath = "/access.log"
@@ -195,7 +203,7 @@ EOF
 [http.services]
   [http.services.dummy-service.loadBalancer]
     [[http.services.dummy-service.loadBalancer.servers]]
-      url = "http://127.0.0.1"  # Dummy service - not used
+      url = "http://0.0.0.0"  # Dummy service - not used
 EOF
         destination = "local/dynamic.toml"
         change_mode = "noop"

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
+      }
+    }
+  }
+}
+

jobs/monitoring/uptime-huma.hcl

@@ -1,44 +0,0 @@
-job "uptime-kuma" {
-  datacenters = ["aperture"]
-  type        = "service"
-
-  group "web" {
-    count = 1
-
-    network {
-      port "http" {
-        to = 3001
-      }
-    }
-
-    service {
-      port = "http"
-
-      check {
-        type     = "http"
-        path     = "/"
-        interval = "10s"
-        timeout  = "2s"
-      }
-
-      tags = [
-        "traefik.enable=true",
-        "traefik.http.routers.uptime-kuma.rule=Host(`status.redbrick.dcu.ie`)",
-        "traefik.http.routers.uptime-kuma.entrypoints=web,websecure",
-        "traefik.http.routers.uptime-kuma.tls.certresolver=lets-encrypt",
-      ]
-    }
-
-    task "web" {
-      driver = "docker"
-
-      config {
-        image = "louislam/uptime-kuma:1"
-        ports = ["http"]
-        volumes = [
-          "/storage/nomad/uptime-kuma/data:/app/data"
-        ]
-      }
-    }
-  }
-}

jobs/services/wiki/LocalSettings.php

@@ -41,7 +41,7 @@ $wgDBpassword = "{{ key "mediawiki/db/password" }}";
 # MySQL specific settings
 $wgDBprefix = "rbwiki_";
 # MySQL table options to use during installation or update
-$wgDBTableOptions = "ENGINE=InnoDB, DEFAULT CHARSET=utf8mb4";
+$wgDBTableOptions = "ENGINE=InnoDB, DEFAULT CHARSET=binary";
 
 ## Shared memory settings
 $wgMainCacheType = CACHE_NONE;
@@ -89,15 +89,11 @@ wfLoadSkin( 'Vector' );
 wfLoadSkin( 'Citizen' );
 wfLoadSkin( 'Timeless' );
 wfLoadSkin( 'MinervaNeue' );
-wfLoadSkin( 'Medik' );
 
 $wgCitizenThemeColor = "#a81e22";
 $wgCitizenShowPageTools = "permission";
 $wgCitizenSearchDescriptionSource = "pagedescription";
 
-$wgMedikColor = "#a81e22";
-$wgMedikShowLogo = "main";
-
 $wgLocalisationUpdateDirectory = "$IP/cache";
 
 # load extensions

jobs/services/wiki/mediawiki-backup.hcl

@@ -30,7 +30,7 @@ job_name=$(echo ${NOMAD_JOB_NAME} | cut -d "/" -f 1)
 
 nomad alloc exec -task rbwiki-db $alloc_id mariadb-dump -u {{ key "mediawiki/db/username" }} -p'{{ key "mediawiki/db/password"}}' {{ key "mediawiki/db/name" }} > "${file}"
 
-find /storage/backups/nomad/wiki/mysql/rbwiki-mysql* -ctime +30 -exec rm {} \; || true
+find /storage/backups/nomad/wiki/mysql/rbwiki-mysql* -ctime +3 -exec rm {} \; || true
 
 if [ -s "$file" ]; then # check if file exists and is not empty
   echo "Backup successful"

jobs/services/wiki/mediawiki.hcl

@@ -213,18 +213,6 @@ EOH
       template {
         data = <<EOH
 [mysqld]
-# Ensure full UTF-8 support
-character-set-server = utf8mb4
-collation-server = utf8mb4_unicode_ci
-skip-character-set-client-handshake
-
-# Fix 1000-byte key length issue
-innodb_large_prefix = 1
-innodb_file_format = Barracuda
-innodb_file_per_table = 1
-innodb_default_row_format = dynamic
-
-# Performance optimizations (Keep these based on your system)
 max_connections = 100
 key_buffer_size = 2G
 query_cache_size = 0
@@ -236,14 +224,13 @@ innodb_io_capacity = 200
 tmp_table_size = 5242K
 max_heap_table_size = 5242K
 innodb_log_buffer_size = 16M
+innodb_file_per_table = 1
 
+bind-address = 0.0.0.0
 # Logging
 slow_query_log = 1
 slow_query_log_file = /var/log/mysql/slow.log
 long_query_time = 1
-
-# Network
-bind-address = 0.0.0.0
 EOH
 
         destination = "local/conf.cnf"

jobs/socs/esports-discord-bot.hcl

@@ -10,7 +10,6 @@ job "esports-discord-bot" {
 
       config {
         image = "ghcr.io/aydenjahola/discord-multipurpose-bot:main"
-        force_pull = true
       }
 
       resources {
@@ -29,15 +28,6 @@ RAPIDAPI_KEY={{ key "socs/esports/bot/rapidapi/key" }}
 TRACKER_API_KEY={{ key "socs/esports/bot/trackerapi/key" }}
 TRACKER_API_URL={{ key "socs/esports/bot/trackerapi/url" }}
 WORDNIK_API_KEY={{key "socs/esports/bot/wordnikapi/key" }}
-HUGGING_FACE_API_KEY={{ key "socs/esports/bot/huggingface/key" }}
-
-RCON_HOST=esports-mc-rcon.service.consul
-
-# https://discuss.hashicorp.com/t/passing-registered-ip-and-port-from-consul-to-env-nomad-job-section/35647
-{{ range service "esports-mc-rcon" }}
-RCON_PORT={{ .Port }}{{ end }}
-
-RCON_PASSWORD={{ key "games/mc/esports-mc/rcon/password" }}
 EOH
         destination = "local/.env"
         env = true

jobs/user-vms/bastion-vm-backup.hcl

@@ -38,7 +38,7 @@ alloc_id=$(nomad job status -verbose bastion-vm | grep running | tail -n 1 | cut
 job_name=$(echo ${NOMAD_JOB_NAME} | cut -d "/" -f 1)
 
 echo "Backing up alloc id: ${alloc_id} on: ${host} to ${path}/${file}..."
-ssh -i {{ key "bastion-vm/service/key" }} {{ key "bastion-vm/service/user" }}@${host} "sudo cat /opt/nomad/alloc/${alloc_id}/bastion-vm/local/bastion-vm.qcow2" > ${path}/${file}
+scp -B -i {{ key "bastion-vm/service/key" }} {{ key "bastion-vm/service/user" }}@${host}:/opt/nomad/alloc/${alloc_id}/bastion-vm/local/bastion-vm.qcow2 ${path}/${file}
 
 find ${path}/bastion-vm-* -ctime +2 -exec rm {} \; || true