Add salt-master and salt-minion alerts
Change-Id: I1407f89306b192ace13752f1fbc0a8594310aead
diff --git a/metadata/service/support.yml b/metadata/service/support.yml
index d61c6da..6472c6c 100644
--- a/metadata/service/support.yml
+++ b/metadata/service/support.yml
@@ -5,6 +5,8 @@
_support:
telegraf:
enabled: true
+ prometheus:
+ enabled: true
collectd:
enabled: false
heka:
diff --git a/salt/meta/prometheus.yml b/salt/meta/prometheus.yml
new file mode 100644
index 0000000..8575ac9
--- /dev/null
+++ b/salt/meta/prometheus.yml
@@ -0,0 +1,32 @@
+{%- if pillar.salt is defined %}
+{%- if pillar.salt.get('master', {}).get('enabled', False) or pillar.salt.get('minion', {}).get('enabled', False) %}
+server:
+ alert:
+ {%- if pillar.salt.get('master', {}).get('enabled', False) %}
+ SaltMasterProcessDown:
+ if: >-
+ procstat_running{process_name="salt-master"} == 0
+ {%- raw %}
+ labels:
+ severity: warning
+ service: salt-master
+ annotations:
+ summary: 'Salt-master service is down'
+ description: 'Salt-master service is down on node {{ $labels.host }}'
+ {%- endraw %}
+ {%- endif %}
+ {%- if pillar.salt.get('minion', {}).get('enabled', False) %}
+ SaltMinionProcessDown:
+ if: >-
+ procstat_running{process_name="salt-minion"} == 0
+ {%- raw %}
+ labels:
+ severity: warning
+ service: salt-minion
+ annotations:
+ summary: 'Salt-minion service is down'
+ description: 'Salt-minion service is down on node {{ $labels.host }}'
+ {%- endraw %}
+ {%- endif %}
+{%- endif %}
+{%- endif %}