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: 75
    heartbeat_retries: 1
    tags: [internal, infrastructure, high]
    restart_threshold: 1
  - name: Vault Cluster
    interval: 75
    heartbeat_retries: 1
    tags: [internal, infrastructure, high]
    restart_threshold: 1
  - name: Etcd Cluster
    interval: 75
    heartbeat_retries: 1
    tags: [internal, database, high]
    restart_threshold: 1
  - name: Patroni Cluster
    interval: 75
    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: 75
    heartbeat_retries: 1
    tags: [internal, gateway, high]
    restart_threshold: 1
  - name: Rabbitmq Cluster
    interval: 75
    heartbeat_retries: 1
    tags: [internal, gateway, medium]
    restart_threshold: 3
  - name: Redis Sentinel
    interval: 75
    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}/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"]