We recently migrated SaaSScout's analysis engine from a pure Serverless architecture (Vercel) to a containerized edge setup (Fly.io).

The Challenge: Maintaining persistent WebSocket connections for real-time status updates in a Lambda-based environment resulted in high latency and frequent connection drops.

The Solution: 1. Migration to Fly.io: Utilizing persistent Node.js instances to handle the Socket.IO handshake and state. 2. Hybrid Transport Strategy: Implementing `transports: ['polling', 'websocket']`. This ensures that the initial connection is established via HTTP long-polling (highly compatible) before upgrading to WebSocket. 3. Edge Optimization: Deploying instances closer to the user to reduce the TCP handshake RTT.

The Result: Analysis feedback latency dropped from ~2s (polling) to <150ms (WebSocket).

We'd love to hear your thoughts on infrastructure trade-offs for real-time SaaS apps.