version: "1"
project: "iklim"
domain:
  base: "iklim.co"
nodes:
  prod:
    service:
      - name: iklim-app-01
        ip: "178.104.210.41"
      - name: iklim-app-02
        ip: "178.105.69.1"
      - name: iklim-app-03
        ip: "178.104.219.3"
    db:
      - name: iklim-db-01
        ip: "159.69.117.158"
      - name: iklim-db-02
        ip: "178.104.219.162"
      - name: iklim-db-03
        ip: "159.69.115.105"
  test:
    service:
      - name: iklim-app-01
        ip: "167.235.194.61"
    db:
      - name: iklim-db-01
        ip: "167.235.205.93"
tags:
  - external
  - internal
  - high
  - medium
  - low
  - database
  - gateway
  - infrastructure
  - observability
notifications:
  slack-high:
    type: slack
    webhook_env: UK_SLACK_WEBHOOK_HIGH
  slack-medium:
    type: slack
    webhook_env: UK_SLACK_WEBHOOK_MEDIUM
  slack-low:
    type: slack
    webhook_env: UK_SLACK_WEBHOOK_LOW
groups:
  - name: "Infrastructure"
    status_page: "iklim-{env}-ops"
    notifications: [slack-high]
    tags: [internal, infrastructure]
    children: [SWARM-CLUSTER, VAULT-CLUSTER, STORAGEBOX-MOUNT, SWAG-TLS]
  - name: "Data Layer"
    status_page: "iklim-{env}-ops"
    notifications: [slack-high]
    tags: [internal, database]
    children: [ETCD-CLUSTER, PATRONI-CLUSTER, MONGODB-REPLICASET]
  - name: "Gateway & Messaging"
    status_page: "iklim-{env}-ops"
    notifications: [slack-high]
    tags: [internal, gateway]
    children: [APISIX-GATEWAY, RABBITMQ-CLUSTER, REDIS-SENTINEL]
  - name: "External Availability - Critical"
    status_page: "iklim-{env}-ops"
    notifications: [slack-high]
    tags: [external, high]
    children: [EXT-HTTPS-API, EXT-DNS-API, EXT-DNS-ROOT, EXT-PING-APP01, EXT-PING-APP02, EXT-PING-APP03]
  - name: "External Availability - General"
    status_page: "iklim-{env}-ops"
    notifications: [slack-medium]
    tags: [external, medium]
    children: [EXT-HTTPS-GRAFANA, EXT-PING-DB01, EXT-PING-DB02, EXT-PING-DB03]
  - name: "Observability"
    status_page: "iklim-{env}-tools"
    notifications: [slack-low]
    tags: [internal, observability]
    children: [PROMETHEUS, GRAFANA, PORTAINER, LOKI, EXT-HTTPS-PORTAINER, EXT-HTTPS-APIGW]
push_monitors:
  - name: SWARM-CLUSTER
    interval: 60
    heartbeat_retries: 1
    tags: [internal, infrastructure, high]
    restart_threshold: 1
  - name: VAULT-CLUSTER
    interval: 60
    heartbeat_retries: 1
    tags: [internal, infrastructure, high]
    restart_threshold: 1
  - name: ETCD-CLUSTER
    interval: 60
    heartbeat_retries: 1
    tags: [internal, database, high]
    restart_threshold: 1
  - name: PATRONI-CLUSTER
    interval: 60
    heartbeat_retries: 1
    tags: [internal, database, high]
    restart_threshold: 1
  - name: MONGODB-REPLICASET
    interval: 120
    heartbeat_retries: 1
    tags: [internal, database, high]
    restart_threshold: 1
  - name: APISIX-GATEWAY
    interval: 60
    heartbeat_retries: 1
    tags: [internal, gateway, high]
    restart_threshold: 1
  - name: RABBITMQ-CLUSTER
    interval: 60
    heartbeat_retries: 1
    tags: [internal, gateway, medium]
    restart_threshold: 3
  - name: REDIS-SENTINEL
    interval: 60
    heartbeat_retries: 1
    tags: [internal, database, medium]
    restart_threshold: 3
  - name: SWAG-TLS
    interval: 3600
    heartbeat_retries: 1
    tags: [internal, infrastructure, medium]
    restart_threshold: 3
  - name: STORAGEBOX-MOUNT
    interval: 300
    heartbeat_retries: 1
    tags: [internal, infrastructure, medium]
    restart_threshold: 1
  - name: PROMETHEUS
    interval: 120
    heartbeat_retries: 1
    tags: [internal, observability, low]
    restart_threshold: 5
  - name: GRAFANA
    interval: 120
    heartbeat_retries: 1
    tags: [internal, observability, low]
    restart_threshold: 5
  - name: PORTAINER
    interval: 120
    heartbeat_retries: 1
    tags: [internal, observability, low]
    restart_threshold: 5
  - name: LOKI
    interval: 120
    heartbeat_retries: 1
    tags: [internal, observability, low]
    restart_threshold: 5
http_monitors:
  - name: EXT-HTTPS-API
    url: "https://api{suffix}.{domain}/actuator/health"
    accepted_statuscodes: ["200"]
    interval: 60
  - name: EXT-HTTPS-GRAFANA
    url: "https://grafana{suffix}.{domain}/api/health"
    accepted_statuscodes: ["200"]
    interval: 60
  - name: EXT-HTTPS-PORTAINER
    url: "https://portainer{suffix}.{domain}"
    accepted_statuscodes: ["200", "401", "403"]
    interval: 120
  - name: EXT-HTTPS-APIGW
    url: "https://apigw{suffix}.{domain}"
    accepted_statuscodes: ["200", "401", "403"]
    interval: 120
dns_monitors:
  - name: EXT-DNS-API
    hostname: "api{suffix}.{domain}"
    dns_resolve_type: A
    interval: 60
  - name: EXT-DNS-ROOT
    hostname: "{domain}"
    dns_resolve_type: A
    interval: 60
ping_monitors:
  interval: 60
  max_retries: 1
status_pages:
  - slug: "iklim-{env}-status"
    title: "iklim.co API Status"
    public: true
    groups: ["External Availability - Critical"]
  - slug: "iklim-{env}-ops"
    title: "iklim.co [{env}] Infrastructure"
    public: false
    groups:
      - "Infrastructure"
      - "Data Layer"
      - "Gateway & Messaging"
      - "External Availability - Critical"
      - "External Availability - General"
  - slug: "iklim-{env}-tools"
    title: "iklim.co [{env}] Tools"
    public: false
    groups: ["Observability"]