プレゼンスとオンライン状態の追跡

Presence is the real-time indication of whether a user is online, away, or offline. Chat apps, collaborative editors, and games all rely on it. Redis is ideal because presence data is ephemeral and high-churn.

A first idea: store a flag per user. The problem is detecting when someone disconnects ungracefully (closed laptop, lost network). A stale flag would mark them online forever.

The robust pattern uses heartbeats. The client refreshes a key with a short TTL every few seconds. If heartbeats stop, the key expires and the user is considered offline.

To list everyone online, keep a set of active users and check membership, or scan presence keys. A sorted set keyed by last-seen timestamp is efficient.

Storing last-seen as the score lets you both list members and prune the stale ones in one query.

• ZADD online <now> user:id on each heartbeat
• ZREMRANGEBYSCORE online -inf <cutoff> removes timed-out users

When a user comes online or drops off, publish an event so other clients update their UI immediately.

Combine the sorted set prune with keyspace expiry. When a presence key expires, a keyevent fires, and a worker publishes a went-offline event to subscribers.

In a chat app you often need presence per room. Use one set per room and add or remove the user as they join or leave.

A user may be online from several devices. Count active connections so the user only goes offline when the last device disconnects.

A complete presence system: heartbeat TTL keys, a last-seen sorted set, connection counters for multi-device, and Pub/Sub events to push changes to clients in real time.

Choose a heartbeat interval shorter than the TTL (for example heartbeat every 10s, TTL 15s) so a single missed beat does not flap a user offline. Shorter intervals mean faster detection but more traffic.

You built a real-time presence system using heartbeat TTL keys, a last-seen sorted set with score-based pruning, per-room presence sets, multi-device connection counters, and Pub/Sub broadcasts for instant status updates. Tune heartbeat and TTL intervals to balance detection speed against traffic.