2021-10-22 23:53:42 +02:00
|
|
|
groups:
|
|
|
|
|
- name: alert.rules
|
|
|
|
|
rules:
|
|
|
|
|
- alert: InstanceDown
|
2021-10-23 18:14:51 +02:00
|
|
|
expr: up{job="node-exporter"} == 0
|
2021-10-23 18:26:08 +02:00
|
|
|
for: 5m
|
2021-10-22 23:53:42 +02:00
|
|
|
labels:
|
|
|
|
|
severity: critical
|
|
|
|
|
annotations:
|
|
|
|
|
summary: "Endpoint {{ $labels.instance }} down"
|
|
|
|
|
description: "{{ $labels.instance }} of job {{ $labels.job }} has been down for more than 1 minutes."
|
2022-09-11 17:01:24 +02:00
|
|
|
- alert: InstanceUptime
|
|
|
|
|
expr: time() - node_boot_time_seconds{job="node-exporter"} >= 7776000
|
|
|
|
|
labels:
|
|
|
|
|
severity: warning
|
|
|
|
|
annotations:
|
|
|
|
|
summary: "Uptime of {{ $labels.instance }} more than 90 days"
|
|
|
|
|
description: "The last reboot of {{ $labels.instance }} was more than 90 days ago"
|
