blob: 557dca77e9807e5a803916e57f8801a99f3f4be6 [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: 'Kibana process is down on node {{ $labels.host }}'
{% endraw %}
KibanaProcessesDownMinor:
if: >-
count(procstat_running{process_name="kibana"} == 0) >= count(procstat_running{process_name="kibana"}) * {{ monitoring.service_failed_warning_threshold_percent }} and count(procstat_running{process_name="kibana"} == 0) < count(procstat_running{process_name="kibana"}) * {{ monitoring.service_failed_critical_threshold_percent }}
{% raw %}
labels:
severity: minor
service: kibana
annotations:
summary: 'Medium percentage of Kibana processes are down'
description: 'More than {%- endraw %} {{monitoring.service_failed_warning_threshold_percent*100}}%{%- raw %} and less than {%- endraw %} {{monitoring.service_failed_critical_threshold_percent*100}}%{%- raw %} of Kibana processes 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: 'High percentage of Kibana processes are down'
description: 'More than {%- endraw %} {{monitoring.service_failed_critical_threshold_percent*100}}%{%- raw %} of Kibana processes 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. Kibana service is not available'
{% endraw %}
{%- endif %}