blob: 2dda0cab592f3c501e38e36ba141841ee3075b70 [file] [log] [blame]
{%- from "kibana/map.jinja" import server, monitoring with context %}
{%- if server.get('enabled', False) %}
server:
alert:
KibanaProcessDown:
if: >-
procstat_running{process_name="kibana"} == 0
{% raw %}
labels:
severity: minor
service: kibana
annotations:
summary: "Kibana process is down"
description: "The Kibana process on the {{ $labels.host }} node is down."
{% endraw %}
KibanaProcessesDownMinor:
if: >-
count(procstat_running{process_name="kibana"} == 0) >= count(procstat_running{process_name="kibana"}) * {{ monitoring.service_failed_warning_threshold_percent }}
{% raw %}
labels:
severity: minor
service: kibana
annotations:
summary: "{%- endraw %}{{monitoring.service_failed_warning_threshold_percent*100}}%{%- raw %} of Kibana processes are down"
description: "{{ $value }} Kibana processes (>= {%- endraw %}{{monitoring.service_failed_warning_threshold_percent*100}}%{%- raw %}) are down."
{% endraw %}
KibanaProcessesDownMajor:
if: >-
count(procstat_running{process_name="kibana"} == 0) >= count(procstat_running{process_name="kibana"}) * {{ monitoring.service_failed_critical_threshold_percent }}
{% raw %}
labels:
severity: major
service: kibana
annotations:
summary: "{%- endraw %}{{monitoring.service_failed_critical_threshold_percent*100}}%{%- raw %} of Kibana processes are down"
description: "{{ $value }} Kibana processes (>= {%- endraw %}{{monitoring.service_failed_critical_threshold_percent*100}}%{%- raw %}) are down."
{% endraw %}
KibanaServiceOutage:
if: >-
count(procstat_running{process_name="kibana"} == 0) == count(procstat_running{process_name="kibana"})
{% raw %}
labels:
severity: critical
service: kibana
annotations:
summary: "Kibana service outage"
description: "All Kibana processes are down."
{% endraw %}
{%- endif %}