The problems now seem mostly related to starting new instances. Our capacity is slowly decaying as existing services spin down and new EC2 workloads fail to start.