AWS Knowledge Hub

English

Tiếng Việt

English

Tiếng Việt

Appearance

Monitoring Strategy ​

A comprehensive observability strategy is essential for maintaining reliable, performant applications on AWS. This guide outlines best practices for monitoring, logging, and tracing across your infrastructure.

The Three Pillars of Observability ​

Metrics ​

What to Monitor:

  • Infrastructure metrics: CPU, memory, disk I/O, network throughput
  • Application metrics: Request rates, error rates, latency (p50, p95, p99)
  • Business metrics: User signups, transactions, revenue
  • Custom metrics: Domain-specific KPIs

Best Practices:

  • Use CloudWatch Metrics for AWS service monitoring
  • Implement custom metrics for application-specific insights
  • Set up metric alarms for proactive alerting
  • Use metric math to create composite metrics
  • Leverage CloudWatch Insights for log-based metrics

Logs ​

What to Log:

  • Application logs: Errors, warnings, info-level events
  • Access logs: API Gateway, ALB, CloudFront access patterns
  • Audit logs: CloudTrail for API calls and security events
  • System logs: OS-level events, container logs

Best Practices:

  • Centralize logs in CloudWatch Logs or S3
  • Use structured logging (JSON format)
  • Implement log retention policies
  • Enable log encryption at rest
  • Use log groups for logical organization
  • Set up log-based alarms

Traces ​

What to Trace:

  • Request flows: End-to-end request paths across services
  • Service dependencies: Inter-service communication patterns
  • Performance bottlenecks: Slow operations and database queries
  • Error propagation: How errors flow through distributed systems

Best Practices:

  • Use AWS X-Ray for distributed tracing
  • Instrument applications with X-Ray SDK
  • Trace all external API calls and database queries
  • Use sampling to control trace volume
  • Create service maps for visualization

Monitoring Architecture Patterns ​

Centralized Monitoring ​

Pattern:

  • Single CloudWatch account for all metrics and logs
  • Cross-account log aggregation
  • Centralized dashboards and alarms

Use Cases:

  • Multi-account AWS organizations
  • Enterprise environments
  • Compliance requirements

Distributed Monitoring ​

Pattern:

  • Service-specific monitoring per account
  • Aggregated views for cross-service insights
  • Independent alerting per service

Use Cases:

  • Microservices architectures
  • Independent service teams
  • High isolation requirements

Alerting Strategy ​

Alert Levels ​

  1. Critical: Immediate action required (P0 incidents)

    • Service outages
    • Security breaches
    • Data loss events

  2. High: Urgent attention needed (P1 incidents)

    • Performance degradation
    • Error rate spikes
    • Capacity thresholds

  3. Medium: Monitor and investigate (P2 incidents)

    • Warning conditions
    • Trend anomalies
    • Resource utilization

  4. Low: Informational (P3 incidents)

    • Scheduled maintenance
    • Capacity planning
    • Optimization opportunities

Alert Best Practices ​

  • Avoid alert fatigue: Only alert on actionable conditions
  • Use composite alarms: Combine multiple metrics to reduce noise
  • Implement alert routing: Route to appropriate teams/channels
  • Set up escalation policies: Define clear escalation paths
  • Test alerting: Regularly verify alerts work correctly
  • Document runbooks: Provide clear response procedures

Cost Optimization ​

Monitoring Costs ​

  • CloudWatch Metrics: First 10,000 custom metrics free, then $0.30/metric/month
  • CloudWatch Logs: $0.50/GB ingested, $0.03/GB stored/month
  • X-Ray: First 100,000 traces free, then $5.00/1M traces

Optimization Strategies ​

  • Use metric filters instead of custom metrics where possible
  • Implement log sampling for high-volume applications
  • Set appropriate log retention periods
  • Use S3 for long-term log archival
  • Leverage X-Ray sampling to control trace volume
  • Delete unused dashboards and alarms

Implementation Checklist ​

  • Define monitoring requirements for each service
  • Set up CloudWatch dashboards for key metrics
  • Configure log aggregation and retention
  • Implement distributed tracing with X-Ray
  • Create alerting rules with appropriate thresholds
  • Document runbooks for common incidents
  • Set up cost budgets for monitoring services
  • Establish on-call rotation and escalation procedures
  • Regular review and optimization of monitoring setup