MongoDB Atlas, despite the documentation and Atlas UI reflecting M10+ as dedicated tier, is not actually dedicating all resources. While CPU and Memory can be considered dedicated, storage is not dedicated until M300+. After working with MongoDB Support, and support working with AWS support, our performance issues were identified by a noisy neighbor who was choking our disk IOPS even though we are running M40_NVME clusters which are supposed to be high performance.

MongoDB should at a minimum highlight this as a consideration for latency sensitive deployments.

A new Atlas alert should be created to identify issues where slow query performance is due to noisy neighbor.