The elastic resource pool is a service-level hybrid resource scheduling strategy provided by EAS. It allows a service deployed in a dedicated resource group to automatically burst new replicas into a pay-as-you-go public resource group when business peaks trigger a scale-out and dedicated resources are insufficient, ensuring the service's auto scaling capability.
Benefits
Break resource group capacity limits: When a dedicated resource group runs out of capacity, service scale-out can automatically use pay-as-you-go instances from the public resource pool. This prevents horizontal auto scaling (based on metrics such as QPS and CPU) from being limited by the number of physical nodes in the dedicated resource group.
Optimize costs: Handle baseline workloads with low-cost prepaid resources, and use on-demand pay-as-you-go resources only during peak demand. This eliminates the need to maintain excess dedicated resources for peak loads, reducing overall resource costs.
How it works
The scheduling logic for the elastic resource pool follows a clear priority order to balance cost and stability.
Scale-out
When a service triggers a scale-out, the EAS scheduler first attempts to create a new instance within the service's dedicated resource group.
If the dedicated resource group lacks the resources to create the instance (such as no available nodes, or not enough remaining CPU, memory, or GPU to meet the instance requirements), the scheduler provisions a new instance in the public resource group using the instance type you specified when configuring the elastic resource pool.
Scale-in
When a service triggers a scale-in, the system terminates instances in the public resource group first.
The system scales in instances in the dedicated resource group only after all elastic instances in the public resource group are terminated. This strategy releases more expensive elastic resources first while retaining your stable, baseline instances.
Prerequisites
You have created a dedicated resource group. For details, see Work with EAS resource groups.
Configuration
Enable during service deployment
Console
-
Log on to the PAI console. Select a region on the top of the page. Then, select the desired workspace and click Elastic Algorithm Service (EAS).
-
Click Deploy Service. In the Custom Model Deployment section, click Custom Deployment.
On the Custom Deployment page, in the Resource Information section, configure the following key parameters. For information about other parameters, see Custom Deployment.
Parameter
Description
Resource Type
Select EAS Resource Group.
Resource Group
Select an existing dedicated resource group.
Elastic Resource Pool
Turn on the Elastic Resource Pool switch and select an Resource Specification.
After you configure this feature, if a service scale-out requires more resources than the dedicated group can provide, new instances are automatically launched on the configured pay-as-you-go public resources and are billed as public resource group instances. During a scale-in, service instances in the public resource group are terminated first.
Click Deploy.
JSON
The key parameters for enabling the elastic resource pool are as follows. For other parameters, see JSON deployment.
resource_burstable: If this parameter is set to true, the elastic resource pool is enabled.
cloud.networking: When you enable an elastic resource pool, use this field to configure the VPC for the service to ensure network availability during elastic scaling.
ImportantWhen you use a dedicated resource group, you can configure a VPC at the resource group level, but a dedicated resource group can be configured only at the service level (by using the cloud.networking field).
cloud.computing: The instance types that can be used for instances that are scaled out to public resources. For more information, see Using public resources.
The following example shows a JSON configuration:
{
"model_path": "http://examplebucket.oss-cn-shanghai.aliyuncs.com/models/lr.pmml",
"name": "test_burstable_service",
"processor": "pmml",
"metadata": {
"instance": 1,
"cpu": 1,
"resource": "eas-r-xxx",
"resource_burstable": true
},
"cloud": {
"computing": {
"instance_type": "ecs.r7.2xlarge"
},
"networking": {
"security_group_id": "sg-uf68iou5an8j7sxd****",
"vswitch_id": "vsw-uf6nji7pzztuoe9i7****"
}
}
}Enable or disable after deployment
Console
-
Log on to the PAI console. Select a region on the top of the page. Then, select the desired workspace and click Elastic Algorithm Service (EAS).
Find your deployed service and click Update in the Actions column.
On the Update Service page, go to the Resource Information section to enable or disable the feature.
To enable auto scaling:
In the Resource Information section, turn on the Elastic Resource Pool switch and configure the instance type for the public resource group.
To disable auto scaling:
In the Resource Information section, turn off the Elastic Resource Pool switch.
Click Update.
CLI
Use the following commands to enable or disable the auto scaling feature for a deployed service. The example uses the Windows 64-bit version of the client.
If you deploy a service in a dedicated resource group without configuring the cloud.networking parameter, instances that are scaled out to the public resource group do not support direct network connections after auto scaling is enabled.
# Enable the elastic resource pool feature for a service.
eascmdwin64.exe modify <service_name> -Dmetadata.resource_burstable=true
# Disable the elastic resource pool feature for a service.
eascmdwin64.exe modify <service_name> -Dmetadata.resource_burstable=falseWhere: <service_name> specifies the name of the deployed service.
This change affects only new service instances; it does not automatically migrate existing ones. For example, if you enable the elastic resource pool while two instances are pending, they will not automatically move to the public resource group. You can restart the instances in the console, after which they will be scheduled to public resources. Similarly, if you disable auto scaling for a service that is already using it, instances running in the public resource group will not automatically move back to the dedicated resource group.
Related topics
To automatically scale instances based on defined metrics, see Horizontal auto scaling.
To automatically scale replicas to a specific count at scheduled times, see Scheduled auto scaling.