Building Real-Time Chat That Holds Up at Scale

Transport choices

Persistent WebSocket connections give the lowest latency and the best typing/presence experience. They also mean connection state to manage, reconnection storms to survive, and a scaling story for fan-out.

Managed real-time services carry most apps a long way. Move to self-hosted infrastructure when the message volume, not the ambition, demands it.

Delivery guarantees

Client-generated message IDs plus server acknowledgment gives idempotent sends and safe retries. Store the ordering key server-side; never trust client clocks for ordering.

Push as the fallback path

Most messages arrive when the recipient is not in the app. Treat push delivery as a first-class path with its own retry logic, quiet hours and per-conversation preferences — not an afterthought bolted on before launch.

Moderation hooks

Run message content through classification before delivery for the highest-risk categories, and asynchronously for everything else. Keep a retention window long enough to investigate reports and short enough to respect your privacy policy.