Alerts

alert: HTTPSDown
expr: probe_success{instance=~"https://.*",job="blackbox",probe="http_2xx"}
  == 0
for: 5m
labels:
  severity: critical
annotations:
  description: The HTTPS endpoint {{ $labels.instance }} is not responding or returning
    non-2xx status codes.
  summary: HTTPS Endpoint Down

alert: SSLCertificateExpiry
expr: probe_ssl_earliest_cert_expiry{instance=~"https://.*",job="blackbox"}
  < time() + 604800
for: 5m
labels:
  severity: warning
annotations:
  description: The SSL certificate for the HTTPS endpoint {{ $labels.instance }} will
    expire within one week.
  summary: SSL Certificate Expiry

alert: HighCPUUsage
expr: 100
  * (1 - avg by(instance) (irate(node_cpu_seconds_total{mode="idle"}[5m])))
  > 80
for: 5m
labels:
  severity: warning
annotations:
  description: '{{ $labels.instance }} has CPU usage over 80% for the last 5 minutes.'
  summary: High CPU usage on {{ $labels.instance }}

alert: HighDiskUsage
expr: 100
  * (node_filesystem_size_bytes - node_filesystem_free_bytes) / node_filesystem_size_bytes
  > 80
for: 5m
labels:
  severity: warning
annotations:
  description: '{{ $labels.instance }} has disk usage over 80% for the last 5 minutes.'
  summary: High Disk usage on {{ $labels.instance }}

alert: HighMemoryUsage
expr: (node_memory_MemTotal_bytes
  - node_memory_MemFree_bytes - node_memory_Buffers_bytes - node_memory_Cached_bytes)
  / node_memory_MemTotal_bytes * 100 > 80
for: 5m
labels:
  severity: warning
annotations:
  description: '{{ $labels.instance }} has memory usage over 80% for the last 5 minutes.'
  summary: High Memory usage on {{ $labels.instance }}