AWS re:Invent 2025 - Best practices for serverless developers (CNS403)

Serverless Best Practices for Developers

Scaling Serverless Applications

• Serverless is the default way to build cloud applications and enables rapid growth
• Followed the journey of Emily, who started with a single Lambda function and scaled to 60 global locations processing 3.8 million orders per year
• Key principles for scaling serverless applications:
  • Right-Sizing Functions: Each Lambda function should have a single clear responsibility, with the right amount of memory allocated for performance and cost optimization
  • Organizing Infrastructure: Use a pragmatic approach for repos, stacks, and services based on current needs, not future theoretical perfection
  • Enabling Team Autonomy: Allow teams to choose runtimes and tooling based on their expertise, aligned with domain-driven organization

Transitioning from Synchronous to Asynchronous

• Synchronous architecture led to cascading failures and poor user experience during peak traffic
• Moved to an event-driven, asynchronous design using services like Event Bridge, SQS, and AppSync
  • Provides immediate response to the client, with asynchronous processing in the background
  • Ensures failures in one system don't break others, and allows independent scaling of services
• Leveraged Lambda's event source mapping for efficient asynchronous processing from various event sources
• Detailed best practices for SQS configuration, including visibility timeout, retry policy, dead-letter queues, and flow control

Reducing Unnecessary Code

• Avoided maintaining unnecessary Lambda functions by using direct service integrations and configuration-as-code
  • Used API Gateway to directly integrate with services like DynamoDB, SQS, and Step Functions
  • Leveraged Event Bridge's built-in transformation and filtering capabilities to replace simple data transformation Lambda functions
  • Utilized Step Functions' direct service integrations to remove many proxy Lambda functions

Optimizing Lambda Performance

• Focused on memory optimization, cold start mitigation, and runtime upgrades to improve performance
  • Leveraged the relationship between memory, CPU, and cost to find the sweet spot for each function
  • Used tools like Lambda Power Tuning to automate the memory optimization process
  • Implemented techniques like lazy initialization, connection reuse, and native compilation to reduce cold starts
  • Upgraded runtimes to take advantage of performance improvements in newer versions

Building Resilient and Observable Systems

• Implemented comprehensive error handling and retry mechanisms, including SQS dead-letter queues and Lambda on-failure destinations
  • Ensured idempotency for operations to handle retries, duplicates, and out-of-order processing
  • Utilized Step Functions' built-in retry and error handling capabilities
• Improved observability by embedding business context in logs and metrics, leveraging tools like AWS CloudWatch and the Power Tools library
  • Enabled rapid troubleshooting and understanding of the business impact of technical issues

Cost Optimization

• Understood the pricing models of serverless services like Lambda and Fargate to make cost-effective architectural decisions
• Optimized logging costs by using structured logging, setting appropriate retention periods, and archiving to S3
• Explored cost-saving strategies like Lambda Power Tuning, event filtering, and compute savings plans

Integrating AI Capabilities

• Recognized that AI is just another workload that can be integrated into a serverless architecture
• Used services like Bedrock Agent Core and MCP Gateway to build AI-powered features, such as a customer service chatbot and personalized recommendations
• Leveraged the serverless principles of event-driven processing, automatic scaling, and pay-per-use pricing to efficiently incorporate AI capabilities

Key Takeaways

• Serverless enables rapid growth and evolution of applications, but requires careful design and optimization
• Asynchronous, event-driven architectures provide resilience, scalability, and better user experience
• Reducing unnecessary code through direct service integrations and configuration-as-code can improve performance and reduce costs
• Comprehensive error handling, observability, and cost optimization are crucial for building reliable and cost-effective serverless systems
• Serverless principles can be applied to integrate AI capabilities, leveraging the benefits of scalability, flexibility, and cost-efficiency

Technical Details

• Lambda function sizing and organization principles
• Event-driven architecture using Event Bridge, SQS, and AppSync
• Lambda's event source mapping for efficient asynchronous processing
• Direct service integrations with API Gateway, Event Bridge, and Step Functions
• Performance optimization techniques like memory optimization, cold start mitigation, and runtime upgrades
• Observability tools like CloudWatch and Power Tools
• Serverless pricing models for Lambda and Fargate
• AI integration using Bedrock Agent Core and MCP Gateway

Business Impact

• Enabled Emily's passion project to grow into a global success, processing 3.8 million orders per year with 35% cost reduction
• Improved customer experience by providing immediate order confirmation and real-time progress updates
• Increased revenue by implementing personalized recommendations and automating customer service
• Reduced operational costs through right-sizing, event filtering, and observability-driven optimizations

Examples

• Transformation from a single monolithic Lambda function to a well-organized, domain-driven serverless architecture
• Migrating from synchronous to asynchronous processing to improve resilience and scalability during traffic spikes
• Replacing numerous proxy Lambda functions with direct service integrations and configuration-as-code
• Implementing comprehensive error handling and observability to enable rapid troubleshooting and understanding of business impact
• Integrating AI-powered features like a customer service chatbot and personalized recommendations using serverless services