Horizontal scaling confirmed: real-time state not held solely in process memory

Storing presence, channel subscriptions, or message history exclusively in process memory means that restarting a single server instance silently resets that state for all users connected to it. In production, auto-scaling and rolling deploys trigger instance restarts routinely — users are dropped from channels, presence shows them offline, and in-flight message queues are lost. ISO 25010 reliability requires that state survival is not contingent on any individual process's uptime.

Externalise real-time state to a shared backend. The Socket.IO Redis adapter handles presence and room membership automatically; use a database or Redis for message history.

import { createAdapter } from '@socket.io/redis-adapter';
import { createClient } from 'redis';

const pub = createClient({ url: process.env.REDIS_URL });
const sub = pub.duplicate();
await Promise.all([pub.connect(), sub.connect()]);

io.adapter(createAdapter(pub, sub));
// Rooms, presence, and subscription state now survive instance restarts

For message history, write to the database on send and query on reconnect — never rely on an in-process buffer as the source of truth. The skip condition applies only when the deployment contract explicitly states single-instance forever.