ceilometer/meta/heka.yml

{%- from "ceilometer/map.jinja" import server with context %}
{%- if server.get('enabled', False) %}
{%- set procs = [] %}
{%- for s in server.services %}
{%- do procs.append('ceilometer-polling' if s == 'ceilometer-agent-central' else s) %}
{%- endfor %}
metric_collector:
  trigger:
    {%- for proc in procs %}
    {{ proc|replace("-", "_") }}:
      description: "There is no {{ proc }} process running."
      severity: down
      rules:
      - metric: process_processes
        field:
          service: {{ proc }}
        relational_operator: '=='
        threshold: 0
        window: 60
        periods: 0
        function: last
    {%- endfor %}
  alarm:
    {%- for proc in procs %}
    {{ proc|replace("-", "_") }}:
      alerting: enabled
      triggers:
      - {{ proc|replace("-", "_") }}
      dimension:
        process: {{ proc }}
    {%- endfor %}
aggregator:
  alarm_cluster:
    {%- for proc in procs %}
    {{ proc|replace("-", "_") }}:
      policy: status_of_members
      alerting: enabled
      group_by: hostname
      match:
        process: {{ proc }}
      members:
      - {{ proc|replace("-", "_") }}
      dimension:
        service: ceilometer-control
        nagios_host: 01-service-clusters
    {%- endfor %}
    ceilometer_control:
      policy: highest_severity
      alerting: enabled_with_notification
      match:
        service: ceilometer-control
      members:
      {%- for proc in procs %}
      - {{ proc|replace("-", "_") }}
      {%- endfor %}
      dimension:
        cluster_name: ceilometer-control
        nagios_host: 00-top-clusters
{%- endif %}