Auto Scaling distributes Elastic Compute Service (ECS) instances evenly across multiple zones, monitors instance health in real time, and automatically replaces failed instances. This topic describes how to deploy a high-availability computing cluster with balanced zone distribution and spot instances to reduce costs.
Scenario
During peak hours for distributed big data computing or AI training, you need to rapidly provision a high-availability computing cluster. For example, an online advertising company running machine learning models for targeted ads requires significant computing resources during high-traffic periods.
Deploying a multi-instance cluster manually introduces the following risks:
Creation failures from lengthy manual processes or insufficient resources in a single zone
Downtime when zone-level instance failures disrupt business operations
Solution
Auto Scaling addresses these risks by automatically deploying ECS instances across multiple zones with a balanced distribution strategy. To further reduce costs, configure spot instances as compute nodes -- they provide the same functionality at a lower price.
Benefits
| Benefit | Description |
|---|---|
| Zero O&M costs | Auto Scaling adjusts ECS instances automatically, eliminating manual intervention. |
| High availability | The balanced distribution policy allocates ECS instances evenly across multiple zones, preventing scaling failures from resource shortages in any single zone. The instance health check feature, enabled by default, keeps all instances in the scaling group available. |
| Cost-effectiveness | Spot instances as compute nodes significantly lower resource costs. |
Prerequisites
Before you begin, make sure that you have:
Assessed your business modules to identify which components need high-availability deployment
A custom image created from your application, to use in the scaling configuration. For details, see Create a custom image from an instance
A VPC with vSwitches in multiple zones
Procedure
Step 1: Create a scaling group
-
Log on to the Auto Scaling console.
In the left-side navigation pane, click Scaling Groups.
Select your region in the top navigation bar.
In the upper-left corner of the Scaling Groups page, click Create.
On the Create by Form tab, configure the following parameters and click Create. Parameters not listed here retain their default values. For more information, see Create scaling groups.
Parameter
Example
Description
Scaling Group Name
test
A name for the scaling group. The name must meet the format requirements displayed on the console.
Type
ECS
Select ECS to create a scaling group that manages ECS instances.
Instance Configuration Source
Create from Scratch
Skip the instance template at this stage. After the scaling group is created, create a scaling configuration separately.
Minimum Number of Instances
100
The minimum number of instances in the scaling group. If the actual count drops below this value, Auto Scaling automatically adds instances until the minimum is reached.
Maximum Number of Instances
120
The maximum number of instances in the scaling group. If the actual count exceeds this value, Auto Scaling removes instances until the count is within the limit.
Scaling Policy (under Show Advanced Settings)
Balanced Distribution Policy
Select Balanced Distribution Policy to distribute instances evenly across zones. We recommend that you configure vSwitches in multiple zones for this policy to take effect.
VPC
vpc-2zeghwzptn5zii0w7\*\*\*\*
The VPC for instances in this scaling group.
vSwitch
vsw-2ze23nqzig8inprou\*\*\*\*, vsw-2zeet2ksvw7f14ryz\*\*\*\*, vsw-2ze94pjtfuj9vaymf\*\*\*\*
Select vSwitches in multiple zones. For better balance and higher scale-out success rates, we recommend that you select multiple vSwitches to evenly distribute ECS instances across zones.
Step 2: Create a scaling configuration
On the Scaling Groups page, find the scaling group and click its ID.
In the upper part of the scaling group details page, click the Scaling Configurations tab.
On the Scaling Configurations tab, click Create Scaling Configuration.
Configure the following parameters and click Create. Parameters not listed here retain their default values. For more information, see Create a scaling configuration of the ECS type.
Section
Parameter
Example
Description
Basic Information
Scaling Configuration Name
test
A name for the scaling configuration.
Billing method
Spot Instance
Auto Scaling itself is free of charge. Charges apply only to ECS instances launched during scale-out events. Select Spot Instance to reduce compute costs. For details, see Billing overview.
Image and Instance
Instance Configuration Mode
Specify Instance Pattern
Select Specify Instance Pattern to configure instance specifications manually.
Instance Attribute Combination
2 vCPUs, 4 GiB of Memory, Enterprise Level
Choose the vCPU and memory combination that fits your workload.
Select Image
test
Select the custom image created from your application for production workloads.
Storage
System Disk
Enterprise SSD (ESSD), 40 GiB, PL0
Select a system disk type and size based on your workload requirements.
Network and Security Group
Security Group
sg-bp18kz60mefsicfg\*\*\*\*
Select an existing security group. To create one, see Create a security group.
Management Settings
Logon Credentials
Set Later
Select Set Later to configure instance passwords after creation.
Step 3: Enable the scaling group
Follow the on-screen instructions to enable the scaling group and scaling configuration.
Result
After the scaling group is enabled with Minimum Number of Instances set to 100, Auto Scaling distributes 100 ECS instances evenly across the specified zones. This maintains high availability for the computing cluster while optimizing resource costs through spot instances.
Auto Scaling automatically handles two failure scenarios:
Insufficient resources in a zone: Auto Scaling deploys instances in other available zones to minimize business impact.
Spot instance reclamation: Auto Scaling replaces reclaimed instances and any unhealthy ECS instances with new ones.
Next steps
If instances become unevenly distributed across zones due to resource shortages, rebalance them. For details, see Rebalance the distribution of ECS instances.