AWS re:Invent 2025 - Drive operational excellence for modern applications (COP327)

Driving Operational Excellence for Modern Applications

Challenges with Modern Applications

• Monolithic applications were hard to scale, had tightly coupled dependencies, and made it difficult for teams to work independently
• Modern applications built on microservices, serverless, and short-lived resources provide more agility and flexibility
• However, the distributed nature of modern applications makes them more complex to observe and troubleshoot

Must-Haves for Operational Excellence

Metrics

• Start with business metrics that indicate application performance and impact to the business
• Use "golden" or "red" metrics like availability, requests, errors, and duration to measure application health
• Add metadata (dimensions/labels) to provide context around metrics
• Leverage tags to query vendor-provided metrics
• Continuously improve by reducing detection and resolution time through "correction of errors" process

Alarms and Notifications

• Use metric insights to create one alarm for many metrics, controlling the scope and data
• Aggregate metrics into a single series for one notification or keep them independent
• Include alarm descriptions with relevant information and links

Distributed Tracing

• Tracing is essential for understanding dependencies and identifying issues in distributed applications
• Use open standards like OpenTelemetry for vendor-neutral, interoperable tracing
• Leverage auto-instrumentation to avoid code changes
• Add manual instrumentation for business-specific context
• Correlate traces with logs and other telemetry
• Control trace volume through sampling techniques like head sampling or tail sampling

Filling Gaps in Telemetry

• Utilize specialized tools like:
  • Synthetic canaries for customer-facing visibility
  • Container Insights for container-based workloads
  • Network Flow Logs for network performance
• Leverage AI and machine learning capabilities:
  • Anomaly detection for logs and metrics
  • Cloudwatch Investigations for automated root cause analysis
  • Amazon Managed Service for Prometheus (AMP) for AI-powered insights

Key Takeaways

• Start with business-relevant metrics and add context through tags and dimensions
• Implement distributed tracing using OpenTelemetry for visibility into modern, distributed applications
• Continuously improve observability by reducing detection and resolution time
• Fill gaps in telemetry using specialized tools and AI/ML capabilities
• Focus on operational excellence to ensure modern applications meet business requirements

Resources

• Observability workshop for hands-on activities
• Observability best practices guide
• Kiosks in the expo village for further discussions