ECS Anywhere

ECS Anywhere extends Amazon ECS to infrastructure outside AWS data centres. The control plane stays in AWS; the compute is yours — on-premises servers, virtual machines, or edge devices.

The point is consistency. The same ECS APIs, CLI commands and console manage containers wherever they run, and the tasks still integrate with CloudWatch for monitoring and IAM for access control. That matters where a workload has to stay on-premises for regulatory reasons, where latency to a data source rules out a Region, or where there is hardware already paid for that is not going anywhere.

It is not a way to make on-premises operations disappear. You still own the hardware, the operating system, the network, capacity and physical security. Compared with ECS on Fargate or EC2, it is the highest-control and highest-overhead option.

Registering an external instance

1. Install the SSM Agent, the ECS agent and a container runtime (Docker) on the host.
2. Register the host as an AWS Systems Manager managed instance — ECS Anywhere depends on Systems Manager for the connection back to AWS.
3. Generate the installation script from the ECS console. It contains the Systems Manager activation key and the commands to install and configure the required software.
4. Run that script on each on-premises virtual machine or bare-metal server.
5. Deploy containers to it using the EXTERNAL launch type in the task definition and service.

The activation key in the script is a credential. Treat the generated script accordingly, and set an activation expiry rather than leaving one open indefinitely.

Compared with EKS Anywhere

ECS Anywhere keeps the control plane in AWS and registers your hardware as capacity. EKS Anywhere is the opposite arrangement: you run the entire Kubernetes cluster, control plane included, on your own infrastructure, using the EKS Distro. ECS Anywhere therefore needs connectivity to AWS to schedule work; EKS Anywhere does not.