haproxy/meta/heka.yml

{%- from "haproxy/map.jinja" import proxy with context %}
{%- if proxy.enabled and proxy.listen is defined and proxy.listen|length > 0 %}

metric_collector:
  trigger:
    haproxy_check:
      description: 'HAProxy cannot be checked'
      severity: down
      rules:
      - metric: haproxy_check
        relational_operator: '=='
        threshold: 0
        window: 60
        periods: 0
        function: last
{%- for listen_name, listen in proxy.listen.iteritems() if listen.get('check', True) %}
    haproxy_{{ listen_name }}_backends_all_down:
      description: 'All API backends are down for {{ listen_name }}'
      severity: down
      rules:
      - metric: haproxy_backend_servers
        field:
          backend: {{ listen_name }}
          state: up
        relational_operator: '=='
        threshold: 0
        window: 60
        periods: 0
        function: last
    haproxy_{{ listen_name }}_backends_majority_down:
      description: 'Less than 50% of backends are up for {{ listen_name }}'
      severity: critical
      rules:
      - metric: haproxy_backend_servers_percent
        field:
          backend: {{ listen_name }}
          state: up
        relational_operator: '<='
        threshold: 50
        window: 60
        periods: 0
        function: last
    haproxy_{{ listen_name }}_backends_one_down:
      description: 'At least one API backend is down for {{ listen_name }}'
      severity: warning
      rules:
      - metric: haproxy_backend_servers
        field:
          backend: {{ listen_name }}
          state: down
        relational_operator: '>'
        threshold: 0
        window: 60
        periods: 0
        function: last
{%- if listen.get('type', None) == 'openstack-service' %}
    haproxy_{{ listen_name }}_http_errors:
      description: 'Too many 5xx HTTP errors have been detected on {{ listen_name }}'
      severity: warning
      rules:
      - metric: haproxy_backend_response_5xx
        field:
          backend: {{ listen_name }}
        relational_operator: '>'
        threshold: 0
        window: 60
        periods: 1
        function: diff
{%- endif %}
{%- endfor %}
  alarm:
    haproxy_check:
      alerting: enabled
      triggers:
      - haproxy_check
      dimension:
        service: haproxy
{%- for listen_name, listen in proxy.listen.iteritems() if listen.get('check', True) %}
    haproxy_{{ listen_name }}_backends:
      alerting: enabled
      triggers:
      - haproxy_{{ listen_name }}_backends_all_down
      - haproxy_{{ listen_name }}_backends_majority_down
      - haproxy_{{ listen_name }}_backends_one_down
      dimension:
        backend: {{ listen_name }}
{%- if listen.get('type', None) == 'openstack-service' %}
    {{ listen_name }}_http_errors:
      triggers:
      - haproxy_{{ listen_name }}_http_errors
      dimension:
        backend: {{ listen_name }}
{%- endif %}
{%- endfor %}
aggregator:
  alarm_cluster:
    haproxy_check:
      policy: majority_of_members
      alerting: enabled_with_notification
      match:
        service: haproxy
      group_by: hostname
      members:
      - haproxy_check
      dimension:
        cluster_name: haproxy-openstack
        nagios_host: 00-top-clusters
{%- for listen_name, listen in proxy.listen.iteritems() if listen.get('check', True) %}
    haproxy_{{ listen_name }}:
      policy: highest_severity
      alerting: enabled
      match:
        backend: {{ listen_name }}
      members:
      - haproxy_{{ listen_name }}_backends
{%- if listen.get('type', None) == 'openstack-service' %}
      - haproxy_{{ listen_name }}_http_errors
{%- endif %}
      dimension:
        service: {{ listen.service_name|default(listen_name) }}
        nagios_host: 01-service-clusters
{%- endfor %}

{%- endif %}