monitoring.yml

@@ -121,8 +121,9 @@
       - prometheus-rives.adm.auro.re
       - prometheus-aurore.adm.auro.re
       - prometheus-ovh.adm.auro.re
+      - prometheus-federate.adm.auro.re
   roles:
-    - prometheus-federate
+    - prometheus_federate
 
 
 # Monitor all hosts

roles/prometheus/tasks/main.yml.save

@@ -0,0 +1,84 @@
+---
+- name: Install Prometheus
+  apt:
+    update_cache: true
+    name:
+      - prometheus
+      - prometheus-snmp-exporter
+  register: apt_result
+  retries: 3
+  until: apt_result is succeeded
+
+- name: Configure Prometheus
+  template:
+    src: prometheus/prometheus.yml.j2
+    dest: /etc/prometheus/prometheus.yml
+    mode: 0644
+  notify: Restart Prometheus
+
+- name: Configure Prometheus alert rules
+  template:
+    src: "prometheus/{{ item }}.j2"
+    dest: "/etc/prometheus/{{ item }}"
+    mode: 0644
+  notify: Restart Prometheus
+  loop:
+    - alert.rules.yml
+    - django.rules.yml
+
+- name: Make Prometheus snmp-exporter listen on localhost only
+  lineinfile:
+    path: /etc/default/prometheus-snmp-exporter
+    regexp: '^ARGS='
+    line: "ARGS=\"--web.listen-address=127.0.0.1:9116\""
+  notify: Restart prometheus-snmp-exporter
+
+# This file store SNMP OIDs
+- name: Configure Prometheus snmp-exporter
+  template:
+    src: "prometheus/snmp.yml.j2"
+    dest: "/etc/prometheus/snmp.yml"
+    mode: 0600
+    owner: prometheus
+  notify: Restart prometheus-snmp-exporter
+
+# We don't need to restart Prometheus when updating nodes
+- name: Configure Prometheus nodes
+  copy:
+    content: "{{ prometheus_targets | to_nice_json }}"
+    dest: /etc/prometheus/targets.json
+    mode: 0644
+
+# We don't need to restart Prometheus when updating nodes
+- name: Configure Prometheus Ubiquity Unifi SNMP devices
+  copy:
+    content: "{{ prometheus_unifi_snmp_targets | to_nice_json }}"
+    dest: /etc/prometheus/targets_unifi_snmp.json
+    mode: 0644
+  when: prometheus_unifi_snmp_targets is defined
+
+- name: Configure Prometheus UPS SNMP devices
+  copy:
+    content: "{{ [{'targets': prometheus_ups_snmp_targets }]7yk[:Cp_g$#dT'yv!. | to_nice_json }}\n"
+    dest: /etc/prometheus/targets_ups_snmp.json
+    mode: 0644
+  when: prometheus_ups_snmp_targets is defined
+
+- name: Configure Prometheus docker monitoring
+  copy:
+    content: "{{ [{'targets': prometheus_docker_targets }] | to_nice_json }}\n"
+    dest: /etc/prometheus/targets_docker.json
+    mode: 0644
+  when: prometheus_docker_targets is defined
+
+- name: Activate prometheus service
+  systemd:
+    name: prometheus
+    enabled: true
+    state: started
+
+- name: Indicate role in motd
+  template:
+    src: update-motd.d/05-service.j2
+    dest: /etc/update-motd.d/05-prometheus
+    mode: 0755

roles/prometheus_federate/templates/prometheus/alert.rules.yml.j2

@@ -13,7 +13,7 @@ groups:
     labels:
       severity: critical
     annotations:
-      summary: "{{ $labels.instance }} est invisible depuis plus de 3 minutes !"
+      summary: "Federate : {{ $labels.exported_instance }} est invisible depuis plus de 3 minutes !"
 
   # Alert for out of memory
   - alert: OutOfMemory
@@ -22,7 +22,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "Mémoire libre de {{ $labels.instance }} à {{ humanize $value }}%."
+      summary: "Federate : Mémoire libre de {{ $labels.exported_instance }} à {{ humanize $value }}%."
 
   # Alert for out of disk space
   - alert: OutOfDiskSpace
@@ -31,7 +31,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "Espace libre de {{ $labels.mountpoint }} sur {{ $labels.instance }} à {{ humanize $value }}%."
+      summary: "Espace libre de {{ $labels.mountpoint }} sur {{ $labels.exported_instance }} à {{ humanize $value }}%."
 
   # Alert for out of inode space on disk
   - alert: OutOfInodes
@@ -40,7 +40,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "Presque plus d'inodes disponibles ({{ $value }}% restant) dans {{ $labels.mountpoint }} sur {{ $labels.instance }}."
+      summary: "Federate : Presque plus d'inodes disponibles ({{ $value }}% restant) dans {{ $labels.mountpoint }} sur {{ $labels.exported_instance }}."
 
   # Alert for high CPU usage
   - alert: CpuUsage
@@ -49,7 +49,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "CPU sur {{ $labels.instance }} à {{ humanize $value }}%."
+      summary: "Federate : CPU sur {{ $labels.exported_instance }} à {{ humanize $value }}%."
 
   # Check systemd unit (> buster)
   - alert: SystemdServiceFailed
@@ -58,7 +58,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "{{ $labels.name }} a échoué sur {{ $labels.instance }}"
+      summary: "Federate : {{ $labels.name }} a échoué sur {{ $labels.exported_instance }}"
 
   # Check UPS
   - alert: UpsOutputSourceChanged
@@ -67,7 +67,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "La source d'alimentation de {{ $labels.instance }} a changé !"
+      summary: "Federate : La source d'alimentation de {{ $labels.exported_instance }} a changé !"
 
   - alert: UpsBatteryStatusWarning
     expr: upsBatteryStatus == 3
@@ -75,7 +75,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "L'état de la batterie de {{ $labels.instance }} est faible !"
+      summary: "Federate : L'état de la batterie de {{ $labels.exported_instance }} est faible !"
 
   - alert: UpsBatteryStatusCritical
     expr: upsBatteryStatus == 4
@@ -83,7 +83,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "L'état de la batterie de {{ $labels.instance }} est affaibli !"
+      summary: "L'état de la batterie de {{ $labels.exported_instance }} est affaibli !"
 
   - alert: UpsHighLoad
     expr: upsOutputPercentLoad > 70
@@ -91,7 +91,7 @@ groups:
     labels:
       severity: critical
     annotations:
-      summary: "La charge de {{ $labels.instance }} est de {{ $value }}% !"    
+      summary: "Federate : La charge de {{ $labels.exported_instance }} est de {{ $value }}% !"
 
   - alert: UpsWrongInputVoltage
     expr: (upsInputVoltage < 210) or (upsInputVoltage > 250)
@@ -99,7 +99,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "La tension d'entrée de {{ $labels.instance }} est de {{ $value }}V."
+      summary: "Federate : La tension d'entrée de {{ $labels.exported_instance }} est de {{ $value }}V."
 
   - alert: UpsWrongOutputVoltage
     expr: (upsOutputVoltage < 220) or (upsOutputVoltage > 240)
@@ -107,7 +107,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "La tension de sortie de {{ $labels.instance }} est de {{ $value }}V."  
+      summary: "Federate : La tension de sortie de {{ $labels.exported_instance }} est de {{ $value }}V."
 
   - alert: UpsTimeRemainingWarning
     expr: upsEstimatedMinutesRemaining < 15
@@ -115,7 +115,7 @@ groups:
     labels:
       severity: warning
     annotations:
-      summary: "L'autonomie restante sur {{ $labels.instance }} est de {{ $value }} min."
+      summary: "Federate : L'autonomie restante sur {{ $labels.exported_instance }} est de {{ $value }} min."
 
   - alert: UpsTimeRemainingCritical
     expr: upsEstimatedMinutesRemaining < 5
@@ -123,7 +123,7 @@ groups:
     labels:
       severity: critical
     annotations:
-      summary: "L'autonomie restante sur {{ $labels.instance }} est de {{ $value }} min."
+      summary: "Federate : L'autonomie restante sur {{ $labels.exported_instance }} est de {{ $value }} min."
 
 
 {% endraw %}

roles/prometheus_federate/templates/prometheus/prometheus.yml.j2

@@ -52,4 +52,5 @@ scrape_configs:
         - '{job="ups_snmp"}'
         - '{job="django"}'
         - '{job="docker"}'
+        - '{job="switch"}'