To associate a listener with an endpoint group, you need to specify the region to which you want to route network traffic. Then, the system routes network traffic to the endpoint in the endpoint group based on the routing type of the listener.
Endpoint groups
Each endpoint group is associated with a region. You can associate an endpoint group with a listener by specifying the region to which you want to distribute network traffic. After you associate an endpoint group with a listener, the system forwards client requests to endpoints in the endpoint group based on the routing type of the listener.
By default, you can create two endpoint groups for a custom routing listener.
Intelligent routing listeners support the following types of endpoint groups:
Default endpoint group: the endpoint group that you configured when you create a listener.
By default, you can create two default endpoint groups for a TCP or UDP listener. If you want to create more endpoint groups, go to the Quota Management page and increase the quota of gaplus_quota_epgs_per_listener. For more information, see Manage GA quotas.
A default endpoint group can be deployed in only one region. You can specify a traffic distribution ratio for each endpoint group. The traffic distribution ratio specifies the proportion of traffic that is distributed to an endpoint group. For more information about how to specify traffic distribution ratios, see Distribute traffic across endpoint groups in different scenarios.
You can create one default endpoint group for each HTTP or HTTPS listener.
Virtual endpoint group: After you create a listener, you can create a virtual endpoint group on the Endpoint Groups page.
By default, you can create 10 virtual endpoint groups for each TCP, HTTP, or HTTPS listener. If you want to create more virtual endpoint groups, go to the Quota Management page and increase the quota named gaplus_quota_vepg_per_listener. For more information, see Manage GA quotas.
UDP listeners do not support virtual endpoint groups.
After you create a virtual endpoint group for a listener, you can create a custom forwarding rule and associate the rule with the virtual endpoint group. After you associate a forwarding rule with an endpoint group, the listener forwards the requests that meet the forwarding conditions to the default endpoint group or the virtual endpoint group specified in the forwarding rule. This way, Global Accelerator (GA) can access multiple endpoints at the same time.
Note
If the TCP listener of your standard GA instance does not support virtual endpoint groups or forwarding rules, the instance may be of an earlier version. Contact your account manager to upgrade your GA instance.
Endpoints
Endpoint types
Endpoints are the destinations of client requests. You can add at most four endpoints to an endpoint group. The following table describes the types of endpoints.
Endpoint deployment | Connection type | Backend service type | Endpoint |
Endpoint deployment | Connection type | Backend service type | Endpoint |
Alibaba Cloud | Private connection | ECS | Elastic Compute Service (ECS) instances in virtual private clouds (VPCs) |
ALB | Application Load Balancer (ALB) instances |
NLB | Network Load Balancer (NLB) instances |
CLB | Internal-facing Classic Load Balancer (CLB) instance in VPCs |
OSS | Object Storage Service (OSS) buckets |
ENI | Elastic network interfaces (ENIs) |
Custom Private IP | Private IP addresses You can add unsupported Alibaba Cloud endpoints or on-premises services to GA instances as private IP addresses. Custom private IP addresses include but are not limited to the following standard private CIDR blocks: 10.0.0.0/8 100.64.0.0/10 172.16.0.0/12 192.168.0.0/16
If you need to access the Internet, the custom private IP address that you use might be determined as the private IP address. If you must use custom private IP address, add routes to the VPC route table. |
Internet connection | Alibaba Cloud public IP address | |
CLB | Internet-facing CLB instances |
Private connection | vSwitch Only custom routing listeners support this type. | vSwitches You can specify the private IP addresses and destination ports of one or more ECS instances in the vSwitch to which you want to forward traffic. |
Outside Alibaba Cloud | Internet connection | Custom Public IP | Custom public IP addresses |
Custom Domain Name | Custom domain names |
Important
If your standard GA instance does not support ECS instances in VPCs, ALB instances, NLB instances, CLB instances in VPCs, ENIs, or custom private IP addresses as endpoints, the GA instance may be using an earlier version. Contact your account manager to upgrade your GA instance.
UDP listeners do not support ALB instances as endpoints.
To ensure that your GA instance is connected to an endpoint, you must configure an access control policy for the endpoint based on the connection type.
Internet connection: The access control policies of the endpoint, such as security group rules or firewall settings, must allow the public IP address of the endpoint.
Private connection: The access policies of the endpoint, such as security group rules or firewall settings, must allow the CIDR block of the vSwitch to which the endpoint belongs. The number of idle private IP addresses of the vSwitch must be equal to or greater than eight.
Custom routing listeners are in invitational preview. To use custom routing listeners, submit an application to your account manager. After your application is approved, you can use custom routing listeners.
Custom routing listeners allow you to specify an IP address and a port in a vSwitch to which you want to forward traffic. You can specify only the IP address and port of an ECS instance.
The public IP address of an endpoint group connected to a GA instance is unique and not shared with other GA instances.
Endpoint weights
You can specify a weight for the endpoint of an intelligent routing listener. The weight determines the ratio of traffic that is distributed to the endpoint. Valid weight values: 0 to 255. Default weight: 255.
GA calculates the sum of the endpoint weights in an endpoint group, and distributes traffic to the endpoints based on the weight proportions of the endpoints.

Proper weight configurations of endpoints can improve traffic distribution, perform stability, response efficiency, and resource unitization. Common scenarios:
Load balancing: Distribute traffic among multiple endpoints in an endpoint group. Endpoints that have higher weights receive more traffic.
Canary release: Gradually migrate traffic to the latest endpoint version during the update of an application by gradually increasing the weight of the endpoint. This facilitates the canary release and result verification.
For example, an application is deployed on Endpoint A that has a weight of 255. The canary release version is deployed on Endpoint B, whose initial weight is 1. In this case, only 1/256th of the traffic is distributed to Endpoint B. After the application passes the test, you can gradually increase the weight of Endpoint B to migrate more traffic from Endpoint A to Endpoint B until traffic is balanced between Endpoint A to Endpoint B. To stop GA from distributing network traffic to Endpoint A, set the weight of Endpoint A to 0.
Note
No traffic is distributed to endpoints whose weight is 0.
If only one endpoint exists in an endpoint group and the weight of the endpoint is between 1 and 255, all traffic destined for the endpoint group is distributed to the endpoint.
If multiple endpoints exist in an endpoint and health checks are disabled, traffic is distributed to all endpoints based on their weights, including unhealthy endpoints. If GA fails to distribute traffic to an unhealthy endpoint multiple times, the traffic is redistributed to other endpoints. This process increases the processing time.
After health checks are enabled, traffic is not distributed to unhealthy endpoints. GA recalculates the weights of the healthy endpoints and distributes traffic among them to ensure service stability and response efficiency.
Health checks
You can enable health checks for endpoint groups of a GA instance. This improves service reliability and availability and prevents service interruptions caused by unhealthy endpoints.
After you enable health checks for an endpoint group, GA periodically checks whether the endpoints are healthy. When GA detects an unhealthy endpoint, GA skips the unhealthy endpoint and distributes requests to healthy endpoints. When the unhealthy endpoint recovers, GA distributes requests to the endpoint again.
Note
You can enable health checks only for endpoint groups of standard GA instances that use intelligent routing listeners.