Skip to main content

ADR-009: Monitoring and Observability Strategy

  • Status: Accepted
  • Date: 2026-02-28
  • Updated: 2026-08-04 (Zero-cost strategy with Sentry Free Developer + CloudWatch Free Tier)

Context​

The project requires unified visibility into frontend/backend failures, performance regressions, and infrastructure health during MVP development and operations. Key constraints and requirements include:

  • Budget constraint: $0/month for monitoring and observability tools
  • Solo developer: One person (tech lead) managing all aspects of the system
  • AWS infrastructure: App Runner, RDS, S3, CloudFront requiring infrastructure monitoring
  • Full-stack visibility: Need to monitor React frontend, FastAPI backend, and PostgreSQL database
  • Developer productivity: Fast error debugging is critical for rapid MVP iteration
  • Operational simplicity: Minimal setup and maintenance overhead for a single-person team

AWS CloudWatch charges for custom metrics beyond 10 free metrics ($0.30/metric/month), while AWS service metrics (App Runner, RDS, S3) are always free. Sentry offers a Free Developer plan for solo developers with full error monitoring, performance tracing, 10 dashboards, and email alerts.

Decision​

Adopt a hybrid monitoring strategy maintaining $0/month cost by leveraging free tiers with clear responsibility boundaries:

Sentry (Free Developer Plan) — Primary application monitoring:

  • Frontend error tracking (JavaScript exceptions, React errors)
  • Backend error tracking (FastAPI exceptions, API errors)
  • Performance monitoring (page loads, API latency, transaction traces)
  • Custom dashboards (10 available)
  • Email alerts for critical errors
  • Release tracking and error correlation

Amazon CloudWatch (Free Tier) — Infrastructure monitoring only:

  • AWS service metrics (App Runner, RDS, S3, CloudFront) — always free
  • Application logs with structured JSON (≤5 GB/month)
  • Critical infrastructure alarms (≤10 alarms)
  • NO custom metrics (avoid $0.30/metric/month charge)

Cost Boundary Rules​

To maintain $0/month:

  1. Sentry: Stay within Free Developer plan limits (one user account, event limits TBD)
  2. CloudWatch: Use only free AWS service metrics; avoid custom metrics
  3. Logs: Keep ingestion under 5 GB/month with 7-day retention
  4. Alarms: Limit to 10 alarms (within free tier)
  5. Application-level metrics: Route to Sentry dashboards, not CloudWatch custom metrics

Consequences​

Positive​

  • Zero monthly cost: Both tools stay within free tiers for MVP scale
  • Best-in-class error debugging: Sentry provides stack traces, user context, breadcrumbs, and session replay
  • Comprehensive infrastructure visibility: CloudWatch native integration with all AWS services (App Runner, RDS, S3, CloudFront)
  • Developer productivity: Rich error context and dashboards accelerate debugging for solo developer
  • No operational overhead: Fully managed services with no self-hosted infrastructure
  • Email alerting: Both tools support email notifications for critical issues
  • Release correlation: Sentry release tagging connects errors to deployments
  • Performance insights: Sentry traces identify slow API endpoints and frontend bottlenecks
  • Scalable foundation: Clear path to paid tiers if MVP grows beyond free tier limits

Negative​

  • Two tools to learn: Developer must understand both Sentry and CloudWatch interfaces
  • Event limit risk: Sentry Free Developer plan has event limits (must manage with sampling/filtering if exceeded)
  • CloudWatch custom metric restriction: Cannot use CloudWatch for application-level custom metrics without incurring $0.30/metric cost
  • Log volume constraint: Must keep CloudWatch log ingestion under 5 GB/month (requires 7-day retention and careful logging)
  • PII scrubbing required: Must configure Sentry to filter sensitive data (emails, tokens, user IDs)
  • No distributed tracing: AWS X-Ray would provide deep tracing but adds complexity and cost
  • CloudWatch query limitations: Logs Insights less powerful than specialized log analysis tools (no advanced regex, limited aggregations)
  • Alert tuning needed: Risk of alert fatigue if thresholds not properly configured
  • Dual monitoring context: Must check both tools during incident response

Alternatives Considered​

  1. CloudWatch only (no Sentry)

    • Considered for simplicity and single-tool approach
    • Not selected because CloudWatch lacks rich error context (no stack traces, user sessions, breadcrumbs) making debugging slower for solo developer

  2. Sentry only (no CloudWatch)

    • Considered for unified application monitoring with excellent developer experience
    • Not selected because Sentry doesn't provide infrastructure metrics (RDS health, App Runner CPU/memory, S3 storage) which are critical for AWS-hosted applications

  3. Paid monitoring services (Datadog, New Relic, Dynatrace)

    • Considered for comprehensive all-in-one APM solution
    • Not selected due to high cost ($75-200+/month) exceeding $0 budget constraint

  4. Self-hosted observability stack (Prometheus + Grafana, ELK stack)

    • Considered for full control and extensibility
    • Not selected due to high operational burden (cluster management, scaling, backups, maintenance) inappropriate for solo developer on MVP timeline

  5. AWS X-Ray for distributed tracing

    • Considered for deep performance insights and request tracing across services
    • Not selected for MVP due to additional SDK integration complexity and costs; can be added post-MVP if tracing needs emerge

  6. CloudWatch with custom metrics (pay $0.30/metric)

    • Considered for unified AWS-native monitoring
    • Not selected because paying for 10-20 custom metrics ($3-6/month) violates $0 budget constraint; Sentry Free provides equivalent metrics at no cost