We experienced roughly 65 minutes of downtime on Tuesday around 5pm PT. While rotating instances in our load balancers, two instances were shut down prior to the newly-provisioned instances being healthy and available. Internal alarms notified the team and the issue was remediated.