AWS us-east-1: a cooling failure in one zone halts Coinbase for several hours

2026-05-07 · Publisher : AWS · Domain : Cloud & DevOps

On 7 May 2026, a thermal event in an availability zone of us-east-1 cuts power to servers. Coinbase halts trading for several hours: three of the five nodes of its matching engine were in the affected zone.

### Background On Thursday 7 May 2026 in the late afternoon (Pacific time), **AWS** reported a thermal event in a single availability zone of its **us-east-1** region (Northern Virginia). Chillers failed, racks were shut down for thermal safety, and AWS saw impaired EC2 instances and degraded EBS volumes. It recommended using the region's other zones during recovery. According to The Stack, it is the fourth us-east-1 incident of 2026. ### Key facts - Coinbase published a post-mortem: its matching engine is a Raft cluster placed in a single AWS cluster placement group, to minimize latency. Three of its five nodes disappeared at the same time: the cluster lost its quorum and could no longer process orders. - Buying, selling, deposits, withdrawals and transfers were impossible for several hours; trading was restored in stages, in cancel-only mode and then through auctions. - Architecture problems identified: no automatic failover to another zone, Kafka infrastructure stranded in the failing zone, recovery requiring emergency code changes and a manual rebuild of the cluster. - Coinbase measures: automated cross-zone recovery, better quorum restoration procedures, a more resilient messaging layer, wider recovery testing. ### Why it matters AWS lost only one zone, which is exactly the scenario multi-zone design is meant to absorb: the outage lasted because the customer had concentrated its critical part in that zone to gain latency. For IT leaders, the lesson is to distinguish what the provider guarantees (a zone can fail) from what the architecture must ensure (survive the loss of a zone), especially for quorum-based systems. ### Takeaway A cooling failure in one zone was enough to halt a financial exchange: resilience is decided in the customer's architecture.

Tags : Incident, AWS, Datacenter, Hyperscaler