Postmortem: DNS race condition took DynamoDB offline in us-east-1

AWS published a postmortem of the October 2025 DynamoDB outage in us-east-1: a race between two automation components deleted the active DNS plan, so the regional endpoint stopped resolving. The failure also blocked EC2 instance launches for about 14 hours.

• DynamoDB maintains hundreds of thousands of DNS records as a scaling primitive
• A delayed Enactor applied a stale plan, then cleanup deleted it as the active one
• DNS was restored by 2:25 AM, but EC2 launches failed until 1:50 PM
• The plan freshness check ran only once, at the start of application