Patterns for Building Event-driven Web & Mobile App Backends • Josh Kahn • GOTO 2024

Event-driven architectures are essential for building modern web and mobile app backends, with three main interaction patterns: real-time events, long-running processes, and push notifications

AWS AppSync (managed GraphQL service) provides a robust solution for real-time updates, offering:

• Built-in WebSocket support
• Flexible authentication options
• Subscription capabilities
• Direct integration with EventBridge

Long-running processes (>30 seconds) require different handling approaches:

• Using Step Functions for event sequencing
• Implementing asynchronous processing
• Storing results in S3 or DynamoDB
• Notifying clients when processing completes

Push notifications can be implemented using:

• Amazon Pinpoint or Firebase for mobile
• Web Push API for browsers
• Requires managing device tokens and user permissions
• Best suited for important updates to avoid notification fatigue

Polling vs Real-time Considerations:

• Polling can be expensive and create unnecessary network traffic
• WebSockets provide more efficient real-time updates
• Consider broadcast vs point-to-point communication needs
• Filter events to reduce costs and complexity

Backend for Frontend (BFF) pattern helps:

• Separate concerns between frontend and backend
• Provide optimized APIs for specific frontends
• Enable better team autonomy
• Support different technology stacks

Architecture patterns should consider:

• Authentication requirements
• Scalability needs
• Development team expertise
• Client platform requirements (web vs mobile)
• Operational complexity

EventBridge serves as an effective backbone for:

• Event choreography
• System integration
• Decoupling services
• Supporting multiple communication patterns

CloudWatch/Logging best practices:

• Use dead letter queues for failed events
• Implement proper error handling
• Monitor connection management
• Debug subscription issues

Performance optimization strategies:

• Filter unnecessary updates
• Implement efficient data mapping
• Use proper batch processing
• Consider partial batch failure handling