Reliability fail: No automated zone failover for Coinbase’s global trading service

This article examines Coinbase's major outage on May 7, where its global trading service went offline for nearly 10 hours due to a regional AWS outage. Coinbase's postmortem reveals that its matching engine was pinned to a single availability zone (AZ) to meet low-latency trading requirements, but the company lacked an automated failover mechanism to another AZ. The article discusses the trade-offs between latency and reliability in distributed systems, the risks of single-AZ dependencies, and contrasts Coinbase's response with other companies that successfully failed over. It provides technical insights into AWS availability zones, regions, and the challenges of distributed consensus in high-frequency trading systems.