Alert Rules

Alert lifecycle

Condition met → Check (N of M) → Alert fired → Channels notified
                                      │
                               Condition clears → Resolved notification

Alerts go through these states:

State	Meaning
Pending	Condition met but not yet for N consecutive checks
Firing	Condition confirmed — notifications sent
Resolved	Condition cleared — resolution sent
Silenced	Firing but suppressed during a maintenance window

Condition types

Threshold

metric operator value for duration

Examples:

cpu.usage_percent > 90 for 3m
mem.used_percent > 85 for 5m
disk./.used_percent > 95 for 1m
response_time_ms > 2000 for 5m

Absence

monitor has no data for 2m  →  agent offline or network partition

Change rate

cpu.usage_percent changes by +30pp in 5m  →  sudden spike
disk.free_bytes drops by 10GB in 1h       →  log storm / disk fill

Alert grouping

Multiple alerts from the same server or domain are grouped into a single notification:

🔴 3 alerts firing on prod-web-01
  • CPU: 94% (threshold 90%)
  • RAM: 88% (threshold 85%)
  • Disk /var: 97% (threshold 95%)

This prevents notification fatigue.

Maintenance windows

Schedule windows during which alerts are silenced:

Every Sunday 02:00–04:00 UTC  — weekly maintenance
2025-06-15 18:00 to 2025-06-15 20:00 UTC — planned deployment

Monitors still run during windows; alerts are suppressed and shown as Silenced.

Resolved notifications are always sent after a maintenance window ends if a condition cleared during it.

Uptime & HTTP Checks Status Pages