prometheus: add quanta alerts

group_vars/prom/prometheus.yml

@@ -40,14 +40,16 @@ prometheus__alert_rules_prometheus:
     labels:
       severity: critical
 
-prometheus__alert_rules_node:
-  - alert: MachineDown
-    expr: "up == 0"
+prometheus__alert_rules_common:
+  - alert: CollectorDown
+    expr: 'up == 0'
     for: 3m
     labels:
       severity: critical
     annotations:
       summary: "Collecteur {{ '$labels.job' | interp }}"
+
+prometheus__alert_rules_node:
   - alert: OutOfMemory
     expr: "( node_memory_MemFree_bytes
              + node_memory_Cached_bytes
@@ -282,16 +284,16 @@ prometheus__alert_rules_quanta:
     labels:
       severity: warning
     annotations:
-      summary: "La température de {{ '$labels.snAgentTempSensorDescr' }} est
-                élevée ({{ '$value' | interp }}°C)"
+      summary: "La température de {{ '$labels.snAgentTempSensorDescr' | interp }}
+                est élevée ({{ '$value' | interp }}°C)"
   - alert: QuantaTemp
     expr: '(snAgentTempValue / 2) > 60'
     for: 0m
     labels:
       severity: critical
     annotations:
-      summary: "La température de {{ '$labels.snAgentTempSensorDescr' }} est
-                très élevée ({{ '$value' | interp }}°C)"
+      summary: "La température de {{ '$labels.snAgentTempSensorDescr' | interp }}
+                est très élevée ({{ '$value' | interp }}°C)"
   - alert: QuantaPowerRedundancyFailure
     expr: 'count by (instance) (snChasPwrSupplyOperStatus{snChasPwrSupplyOperStatus="normal"}) < 2'
     for: 0m
@@ -299,6 +301,7 @@ prometheus__alert_rules_quanta:
       severity: warning
 
 prometheus__alert_rules:
+  common: "{{ prometheus__alert_rules_common }}"
   prometheus: "{{ prometheus__alert_rules_prometheus }}"
   node: "{{ prometheus__alert_rules_node }}"
   keepalived: "{{ prometheus__alert_rules_keepalived }}"