This is what the AWS us-east-1 outage looked like on our graphs. Despite its catastrophic and rare nature, Vercel was still able to successfully serve billions of requests. But let me be clear: our goal is for zero requests to fail. I'm really sorry to our customers and developers for this unacceptable degradation in availability. We’ve designed the Vercel platform with multiple layers of redundancy. Functions execute in multiple availability zones and support failover regions. There are multiple global tiers of metadata replication. Multiple tiers of caches in our CDN that shield the access to origins (e.g.: us-east-1). But there's more to do. The silver lining: this is as good of a chaos engineering exercise as it gets. A multi-service failure of this nature is exceedingly rare (85 impacted services), especially with a global service failure (IAM). Looking forward to sharing our plans for further enhancement of our global resilience and a comprehensive post-mortem.
