AWS re:Invent 2025 - Designing mission critical applications with serverless services (CNS362)

Modernizing Mission-Critical Applications with AWS Serverless

Distributed Systems as Living, Evolving Entities

• Distributed systems are not static, but rather living, evolving entities that require ongoing attention and care
• Key goals of embracing distributed systems:
  • Organizational scalability: Ability to plug in new teams or partners to accelerate delivery
  • Business agility: Ability to quickly adapt to internal or external changes
  • Faster feedback loops: Leverage metrics like DORA to succeed in a complex world
  • Reduced external dependencies: Decrease cognitive load and blast radius

Achieving Modularity Through Serverless

• Modularity can be expressed through code (e.g., encapsulation, design patterns) but requires discipline
• Modularity can also be achieved at the infrastructure level using serverless:
  • Ability to leverage synchronous and asynchronous programming as needed
  • Configuration-driven rather than code-driven, leveraging built-in serverless behaviors
  • Serverless enables focusing on business logic and value generation, not undifferentiated heavy lifting

Booking.com's Serverless Modernization Journey

Challenges with the Legacy Monolithic System

• Slow deployments (monthly instead of weekly)
• Lengthy testing cycles (tens of minutes)
• Frequent rollbacks due to issues
• Tight coupling and cross-dependencies leading to a "big ball of mud"
• Unclear ownership and domain boundaries
• Performance and scalability issues with on-premises infrastructure

Designing a Serverless Target Architecture

• Prioritized goals: isolating unknown code, defining boundaries, removing coupling
• Chose AWS serverless services (Lambda, Step Functions) to achieve modularity and observability
• Addressed team skepticism about serverless through a proof-of-concept and gradual adoption

Key Architectural Decisions and Patterns

• Leveraged Step Functions to orchestrate reusable Lambda functions into clear, isolated workflows
• Used DynamoDB to temporarily store requests, with TTL to manage database size
• Employed SNS and SQS for asynchronous, decoupled communication between components

Overcoming Challenges

• Addressed Lambda cold starts and latency through provisioned concurrency and code optimizations
• Developed a solution for creating personal AWS environments for each team
• Worked with compliance team to address concerns about serverless in a mission-critical system

Testing and Deployment Approach

• Utilized shadow traffic to identify functionality gaps and performance issues
• Adopted an experimentation-based go-live strategy, gradually increasing traffic to the new system
• Leveraged data science to determine optimal traffic division across payment timings

Key Benefits and Outcomes

• Reduced onboarding time for new developers (from 4 months to 1 month)
• Improved developer experience with faster debugging and incident response
• Significantly reduced time-to-market for new features and capabilities
• Enabled 100% of new reservations to be processed through the modernized serverless system

Serverless Migration Patterns and Strategies

• Seven R's of migration: Retain, Retire, Relocate, Rehost, Replatform, Repurchase, Refactor
• Importance of addressing the triangle of people, processes, and technology
• Examples of modernization patterns:
  • Strangler Fig Pattern: Incremental modernization, routing traffic between legacy and new systems
  • Branch by Abstraction: Abstracting complexity of dual implementations
  • Decompose by Business Capability: Aligning architectural boundaries with business functions
  • Decompose by Subdomain: Tailoring logic based on specific business requirements (e.g., regional)
  • Decompose by Transactions: Decoupling user experiences based on immediate vs. long-running actions

Rethinking Communication Patterns with Serverless

• Challenges with synchronous communication in high-load scenarios (e.g., gift code validation)
• Leveraging asynchronous patterns with Amazon DynamoDB, Amazon EventBridge, and AWS Lambda
  • Immediate synchronous feedback loop for customers
  • Asynchronous event propagation and processing for external validations and enrichment
• Embracing eventual consistency over immediate consistency when the business context allows

Key Takeaways

• Distributed systems are living, evolving entities that require ongoing attention and care
• Serverless enables achieving modularity at the infrastructure level, allowing teams to focus on business value
• Booking.com's serverless modernization journey demonstrates the benefits of increased agility, developer productivity, and time-to-market
• Serverless migration patterns and strategies (e.g., Strangler Fig, Decompose by Capability) can help organizations modernize incrementally
• Rethinking communication patterns with asynchronous serverless services can unlock improved user experiences and system resilience