Availability Monitoring & Alerting: 6 Key Concepts You Need to Know

The value of monitoring isn't in whether it can alert — it's in being **accurate, reachable, and locatable**. Clarify these six concepts before choosing tools, …

Insights:Availability Monitoring & Alerting: 6 Key Concepts You Need to Know

Concept 1: Detection Methods

  • External Probing: Simulates user visits from the public internet to verify real-world reachability.
  • Internal Collection: Reads metrics from inside the server to monitor resource and process status.
  • Synthetic Transactions: Walks through full workflows (login, checkout) to assess business availability.

External probing answers "Can users get in?", while internal collection tracks "What's happening inside?" — both are needed.

Concept 2: Availability Metrics

Metric Meaning Common Threshold
Availability Percentage of time the service is accessible over a period 99.9% monthly
Response Time Time to first byte or full response P95 under 2 seconds
Error Rate Proportion of 5xx responses Under 0.5%
Recovery Time Time from alert to restoration Under 30 minutes

Set thresholds based on your own business needs — don't copy others' numbers.

Concept 3: Thresholds and Judgment

Alerting on every single failure is a major source of noise. Common practices:

  • Consecutive N failures before declaring an incident, filtering out transient blips.
  • Multi-region probing — require simultaneous failures across regions to rule out single-route issues.
  • Tiered judgment: Check the entry point first, then the application, and finally dependencies.

Concept 4: Alert Channels

  • Instant channels (IM, SMS) are for incidents that must be handled immediately.
  • Summary channels (daily email reports) are for trends and inspection results.
  • A single incident triggers only one alert; a recovery notification is sent when the issue is resolved.
  • More alert channels do not mean greater reliability—they can actually cause alerts to be overlooked.

    Concept 5: On-Call and Escalation

    • Clearly designate a primary responder and a backup.
    • Define escalation windows: automatic escalation if no response within 15 minutes.
    • Record the start time, diagnostic time, and restoration time for every handling instance.

    An unacknowledged alert is as good as no alert at all.

    Concept 6: Post-Incident Review

    • For every P1 incident, produce a one-page summary: impact scope, root cause, temporary mitigation, and long-term fix.
    • Turn recurring incidents into monitoring rules instead of relying on human vigilance.
    • Clean up long-idle alert rules once per quarter.

    Common Pitfalls

    • Monitoring only the homepage but not critical business pages.
    • Checking only HTTP status codes without verifying whether content is correct.
    • Routing alerts to a group chat with no assigned owner.
    • Setting monitoring frequency too high, effectively DDoSing yourself.

    Boundaries

    Monitoring shortens discovery time but cannot eliminate incidents. Its reasonable expectation is "earlier detection and faster location," not "zero issues."

    Related Reading: