Update: Autoscaling is now GA for all dedicated clusters in MongoDB Atlas
You can now opt in to our compute auto-scaling open beta for any dedicated cluster (M10 and higher) on MongoDB Atlas. When enabled, Atlas will track key resource utilization metrics in real-time and adjust your cluster size up or down as needed, using predictive modeling and proven practices developed from managing tens of thousands of MongoDB deployments.
Predictive Resource Modeling
The open beta for compute auto-scaling is based on a predictive model that uses heuristics from managing numerous MongoDB deployments. Specifically, Atlas will examine two key resource utilization metrics—CPU and memory—every five minutes and determine whether a scaling event is necessary based on certain thresholds. For upscaling events, Atlas will look at the average utilization of each metric over the past hour. For downscaling events, Atlas will track the maximum utilization for both metrics over the past 72 hours. In future iterations, we will improve the algorithm to be more sophisticated and responsive to changes in workload.
Optimize Capacity Management
You can enable compute auto-scaling through either the UI or the public API. Auto-scaling is available on all clusters M10 and higher on Azure and GCP, and on all "General" class clusters M10 and higher on AWS. To enable auto-scaling from the UI, select the Auto-scale “Cluster tier” option and choose a maximum cluster size from the available options.
You can also opt in to automated downscaling, which will require you to also select a minimum cluster size. Opting into downscaling allows you to intelligently save on costs by downsizing your database during periods of low traffic or load.
At this time, compute auto-scaling is not supported by clusters on the M0, M2, or M5 tiers, sharded clusters, clusters with low-CPU or NVMe storage, or clusters with IOPS provisioned.
Set And Forget It
Scaling cluster sizes in Atlas is already fairly easy with the UI or API, but compute auto-scaling takes it one step further and automates the actual operation without any manual intervention required. This is especially useful for developers and DevOps engineers who frequently need to adjust the cluster size to accommodate changes in workload. While the beta feature is currently not suitable for applications that experience sudden spikes, it’s a great fit for those that have regular scaling events, seasonal business patterns, or anticipate steady growth over time.
Atlas automatically applies updates to your cluster in a rolling fashion so there’s no impact to your application. You’ll also receive an email notification any time your cluster has successfully scaled up or down, and you can review all automated scaling operations in your project’s activity feed in the Atlas UI.
An Autonomous Cloud Database
Whether you’re building a new application or supporting an existing one, Atlas helps automate the time-consuming administration tasks so you can free up time to focus on development. Together with other features like storage auto-scaling, self-healing clusters, and automated patches and bug fixes, the new compute auto-scaling makes it easier than ever to run your database in the cloud.
Questions or feedback on the compute auto-scaling open beta in Atlas? Let us know here.