prometheus: monitor iLO resilient mem and battery

roles/prometheus/templates/ilo.rules.yml.j2

@@ -10,6 +10,16 @@ groups:
   - name: ilo.rules
     rules:
 
+      - alert: IloResilientMemoryDegraded
+        expr: cpqHeResilientMemCondition{cpqHeResilientMemCondition!~"ok|other"} == 1
+        for: 3m
+        labels:
+          severity: warning
+        annotations:
+          summary: >-
+            La mémoire vive n'est plus résilente
+            ({{ raw('$labels.cpqHeResilientMemCondition') }})
+
       - alert: IloBiosSelfTestDegraded
         expr: cpqHeHWBiosCondition{cpqHeHWBiosCondition!~"ok|other"} == 1
         for: 3m
@@ -20,6 +30,16 @@ groups:
             Une erreur a été détectée lors du POST du serveur
             ({{ raw('$labels.cpqHeHWBiosCondition') }})
 
+      - alert: IloBatteryDegraded
+        expr: cpqHeSysBatteryCondition{cpqHeSysBatteryCondition!~"ok|other"} == 1
+        for: 3m
+        labels:
+          severity: warning
+        annotations:
+          summary: >-
+            La batterie est dégradée
+            ({{ raw('$labels.cpqHeSysBatteryCondition') }})
+
       - alert: IloTemperatureSensorDegraded
         expr: cpqHeTemperatureCondition{cpqHeTemperatureCondition!~"ok|other"} == 1
         for: 3m

roles/prometheus/templates/snmp.yml.j2

@@ -484,13 +484,28 @@ apc_pdu:
 
 ilo:
   walk:
-  - 1.3.6.1.4.1.232.6.2.15 # Power meter
-  - 1.3.6.1.4.1.232.6.2.16 # POST tests
-  - 1.3.6.1.4.1.232.6.2.6.8.1 # Temperature sensors
-  - 1.3.6.1.4.1.232.6.2.6.7.1 # Fans
-  - 1.3.6.1.4.1.232.6.2.9.3.1 # Power supply
+  - 1.3.6.1.4.1.232.6.2.14.4 # Resilient memory
+  - 1.3.6.1.4.1.232.6.2.15.3 # Power meter
+  - 1.3.6.1.4.1.232.6.2.16.1 # POST tests
+  - 1.3.6.1.4.1.232.6.2.17.1 # Battery
+  - 1.3.6.1.4.1.232.6.2.6.8.1.3 # Temperature sensors location
+  - 1.3.6.1.4.1.232.6.2.6.8.1.4 # Temperature sensors value
+  - 1.3.6.1.4.1.232.6.2.6.8.1.5 # Temperature sensors limit
+  - 1.3.6.1.4.1.232.6.2.6.8.1.6 # Temperature sensors condition
+  - 1.3.6.1.4.1.232.6.2.6.7.1.3 # Fans location
+  - 1.3.6.1.4.1.232.6.2.6.7.1.9 # Fans condition
+  - 1.3.6.1.4.1.232.6.2.9.3.1.5 # Power supply
   - 1.3.6.1.4.1.232.9.2.2 # iLO
   metrics:
+  - name: cpqHeResilientMemCondition
+    oid: 1.3.6.1.4.1.232.6.2.14.4
+    type: EnumAsStateSet
+    help: The resilient memory condition - 1.3.6.1.4.1.232.6.2.14.4
+    enum_values:
+      1: other
+      2: ok
+      3: degraded
+      4: failed
   - name: cpqHePowerMeterCurrReading
     oid: 1.3.6.1.4.1.232.6.2.15.3
     type: gauge
@@ -505,6 +520,20 @@ ilo:
       2: ok
       3: degraded
       4: failed
+  - name: cpqHeSysBatteryCondition
+    oid: 1.3.6.1.4.1.232.6.2.17.1
+    type: EnumAsStateSet
+    help: The battery condition - 1.3.6.1.4.1.232.6.2.17.1
+    indexes:
+    - labelname: cpqHeSysBatteryChassis
+      type: gauge
+    - labelname: cpqHeSysBatteryIndex
+      type: gauge
+    enum_values:
+      1: other
+      2: ok
+      3: degraded
+      4: failed
   - name: cpqHeTemperatureLocale
     oid: 1.3.6.1.4.1.232.6.2.6.8.1.3
     type: EnumAsInfo