Guidelines and restrictions for custom routing accelerators
When you create and work with custom routing accelerators in AWS Global Accelerator, keep the following guidelines and restrictions in mind.
- Supported endpoint destinations
The virtual public cloud (VPC) subnet endpoints in a custom routing accelerator can only include EC2 instances. No other resources, such as load balancers, are supported for custom routing accelerators. The types of EC2 instances that are supported with Global Accelerator are listed in Endpoints for standard accelerators in AWS Global Accelerator.
With custom routing accelerators, Global Accelerator can only route traffic to private IP endpoints on Amazon EC2 instances on VPC subnets. However, gaming customers who want to use custom routing might need to connect to stateful sessions. To do this, the customers run their game servers on Amazon Elastic Kubernetes Service (EKS), with sessions hosted on a specific container running inside a Kubernetes Pod.
To use custom routing in this scenario, you can configure a VPC-CNI plugin to send traffic to Kubernetes Pods through an elastic network interface (ENI) that Global Accelerator creates for each subnet where an endpoint is present. This is a way to use a custom routing accelerator with EKS. The same configuration works to use a custom routing accelerator with Amazon Elastic Container Service (ECS). To learn more, see the detailed steps provided in the following blog post: AWS Global Accelerator Custom Routing with Amazon Elastic Kubernetes Service
. - Port mappings
When you add a VPC subnet, Global Accelerator creates a static port mapping of listener port ranges to the port ranges supported by the subnet. The port mapping for a specific subnet never changes.
You can view the port mapping list for a custom routing accelerator programmatically. For more information, see
ListCustomRoutingPortMappings.- VPC subnet size
VPC subnets that you add to a custom routing accelerator must be a minimum of /28 and a maximum of /17.
- IP address type
Custom routing accelerators support only the IPv4 IP address type.
- Listener port ranges
You must specify enough listener ports, by specifying listener port ranges, to accommodate the number of destinations included in the subnets that you plan to add to your custom routing accelerator. The range that you specify when you create a listener determines how many listener port and destination IP address combinations that you can use with your custom routing accelerator. For maximum flexibility and to reduce the possibility of getting an error that you don't have enough listener ports available, we recommend that you specify a large port range.
Global Accelerator allocates port ranges in blocks when you add a subnet to a custom routing accelerator. We recommend that you allocate listener port ranges linearly and make the ranges large enough to support the number of destination ports that you intend to have. That is, the number of ports you should allocate should be at least the subnet size times the number of destination ports and protocols (destination configurations) that you will have in the subnet.
Note
The algorithm that Global Accelerator uses to allocate port mappings might require you to add more listener ports, beyond this total.
After you create a listener, you can edit it to add additional port ranges and associated protocols, but you can't decrease existing port ranges. For example, if you have a listener port range of 5,000–10,000, you can't change the port range to be 5900–10,000 and you can't change the port range to be 5,000–9,900.
Each listener port range must include a minimum of 16 ports. Listeners support ports 1-65535.
- Destination port ranges
There are two places that you specify port ranges for a custom routing accelerator: the port ranges that you specify when you add a listener and the destination port ranges and protocols that you specify for an endpoint group.
Listener port ranges: The listener ports on the Global Accelerator static IP addresses that your clients connect to. Global Accelerator maps each port to a unique destination IP address and port on a VPC subnet behind the accelerator.
Destination port ranges: The sets of destination port ranges that you specify for an endpoint group (also called the destination configurations) are the EC2 instance ports that receive traffic. To receive traffic on destination ports, the Security Groups associated with your EC2 instances must permit traffic on them.
- Health checks and failover
Global Accelerator does not perform health checks for custom routing accelerators and does not failover to healthy endpoints. Traffic for custom routing accelerators is routed deterministically, regardless of the health of a destination resource.
- All traffic is denied by default
By default, traffic directed through a custom routing accelerator is denied to all destinations in your subnet. To enable destination instances to receive traffic, you must specifically allow all traffic to the subnet or, alternatively, allow traffic to specific instance IP addresses and ports in the subnet.
Updating a subnet or specific destination to allow or deny traffic takes time to propagate across the internet. To determine if a change has propagated, you can call the
DescribeCustomRoutingAcceleratorAPI action to check the accelerator status. For more information, see DescribeCustomRoutingAccelerator.- AWS CloudFormation is not supported
AWS CloudFormation is not supported for custom routing accelerators.
