roles/prometheus/tasks/main.yml

@@ -42,6 +42,7 @@
     mode: u=r,g=r,o=
   loop:
     - server.rules.yml
+    - docker.rules.yml
     - django.rules.yml
     - ups.rules.yml
     - postgres.rules.yml

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

@@ -0,0 +1,50 @@
+---
+{{ ansible_managed | comment }}
+
+{% macro raw(string) -%}
+{{ "{{" }} {{ string }} {{ "}}" }}
+{%- endmacro %}
+
+groups:
+
+  - name: docker.rules
+    rules:
+
+  - alert: ContainerDown
+    expr: docker_container_running_state
+    for: 0m
+    labels:
+      severity: critical
+    annotations:
+      summary: >-
+        Container Docker mort 
+        (instance {{ raw('$labels.instance') }}, container {{ raw('$labels.name') }})
+
+  - alert: ContainerFailed
+    expr: sum(increase(docker_container_restart_count[5m])) > 2
+    for: 0m
+    labels:
+      severity: critical
+    annotations:
+      summary: >- 
+        Container Docker redémarre souvent 
+        (instance raw('{{ $labels.instance') }}, container {{ raw('$labels.name') }})
+
+  - alert: ContainerFailed
+    expr: 
+        (
+          docker_container_cpu_used_total 
+          / 
+          docker_container_cpu_capacity_total
+        ) * 100
+        > 30
+    for: 0m
+    labels:
+      severity: critical
+    annotations:
+      summary: >-
+        Container Docker utilise beaucoup de CPU 
+        (instance {{ raw('$labels.instance') }}, container {{ raw('$labels.name') }}, 
+        value {{ raw('$value | printf "%.1f"'') }})
+
+...