apache/meta/prometheus.yml

{%- from "apache/map.jinja" import server with context %}

{%- if server.get('enabled', False) %}

server:
  alert:
    ApacheServiceDown:
      if: >-
        apache_up != 1
      {%- raw %}
      labels:
        severity: minor
        service: apache
      annotations:
        summary: "Apache service is down"
        description: "The Apache service on the {{ $labels.host }} node is down."
      {%- endraw %}
    ApacheServiceOutage:
      if: >-
        count(label_replace(apache_up, "cluster", "$1", "host", "([^0-9]+).+")) by (cluster) == count(label_replace(apache_up == 0, "cluster", "$1", "host", "([^0-9]+).+")) by (cluster)
      {% raw %}
      labels:
        severity: critical
        service: apache
      annotations:
        summary: "Apache service outage"
        description: "The Apache services within the {{ $labels.cluster }} cluster are down."
      {% endraw %}
    ApacheWorkersAbsent:
      if: >-
        apache_IdleWorkers == 0
      {%- raw %}
      for: 2m
      labels:
        severity: minor
        service: apache
      annotations:
        summary: "Apache has no available idle workers"
        description: "The Apache service on the {{ $labels.host }} node has no available workers for 2 minutes."
      {%- endraw %}
{%- endif %}