Gossip Protocols for Rate Limit Sync
You'll learn how nodes can share rate limit state through eventual consistency without needing a central coordinator like Redis.
The Core Idea
In a gossip protocol, each rate limiter node periodically shares its local counter state with a few random peers, who then share with their peers, and so on — like rumors spreading through a crowd. Over time, every node learns approximate global state without central coordination.
When a request hits Node A, it checks its local counter. Periodically (say, every 100ms), Node A picks 2-3 random peers and exchanges counter values: "I've seen 45 requests for user X." The peer responds: "I've seen 30." They merge this information (typically summing or taking max values) and update their local views.
How It Works for Rate Limiting
Each node maintains:
- Local counters for rate limit keys (user IDs, API keys, etc.)
- A gossip interval timer
- A list of peer nodes
When gossip fires:
- Node selects random peers
- Sends its counter deltas since last sync
- Receives peer deltas
- Merges data (sum counters, reconcile timestamps)
- Makes rate limit decisions based on merged state
The Tradeoffs
Pros:
- No single point of failure (unlike centralized Redis)
- Scales horizontally as nodes join
- Tolerates network partitions
Cons:
- Eventually consistent (not strongly consistent)
- May allow brief over-limit bursts during propagation
- Requires more memory (each node stores state)
- Network chattiness increases with cluster size
Key Takeaway: Gossip protocols enable distributed rate limiting through peer-to-peer state sharing, trading strong consistency for resilience and eliminating central bottlenecks — ideal when approximate limits are acceptable.