Strong Consistency in Practice
You'll learn: How real-world distributed systems like Spanner, etcd, and ZooKeeper implement strong consistency and what performance costs they accept.
From Theory to Production
We've discussed strong consistency models like linearizability and strict serializability, but how do actual production systems deliver these guarantees? Real systems demonstrate that strong consistency is achievable — but comes with measurable performance tradeoffs.
Real Systems That Deliver Strong Consistency
Google Spanner provides strict serializability across globally distributed data centers. It uses synchronized atomic clocks (TrueTime) and consensus protocols to guarantee that every transaction sees a consistent snapshot. The tradeoff? Higher latency for distributed transactions, especially cross-region writes.
etcd is a distributed key-value store used by Kubernetes for cluster coordination. It uses the Raft consensus algorithm to maintain strong consistency across replicas. Every write goes through the leader and requires majority acknowledgment. The cost? Write throughput is limited by the consensus round-trip time.
Apache ZooKeeper coordinates distributed systems with linearizable operations. It guarantees that all clients see operations in the same order. However, achieving this requires synchronous replication, which adds latency compared to eventually consistent systems.
The Common Pattern
These systems share a design: they prioritize correctness over raw speed. They accept:
- Higher write latency (consensus coordination)
- Limited write throughput (single-leader bottleneck)
- Reduced availability during partitions (CP in CAP terms)
In exchange, they eliminate entire classes of bugs related to stale reads, race conditions, and data conflicts.
Key Takeaway: Production systems like Spanner, etcd, and ZooKeeper prove strong consistency is practical for critical operations like coordination, configuration, and financial transactions — as long as you accept the performance costs of consensus and synchronous replication.