Function Compute:Auto scaling for on-demand and provisioned modes

Instance scaling limits

Scaling limits for on-demand instances

By default, the preceding scaling limits are shared by all functions under an Alibaba Cloud account in the same region. To limit the number of instances for a specific function, you can configure function-level scaling control for on-demand instances. After configuration, if the total number of running instances for this function exceeds the limit, Function Compute returns a throttling error.

Scaling limits for provisioned instances

Regional scaling speed limits

Auto scaling rules

Create an auto scaling rule

1. Log on to the Function Compute console. In the left-side navigation pane, click Services & Functions.

2. In the top navigation bar, select a region. On the Services page, click the desired service.

3. On the Functions page, click the function that you want to modify.

4. On the Function Details page, click the Auto Scaling tab and then click Create Rule.

5. On the Create Auto Scaling Rule page, configure the relevant parameters and click Create.

   • Configure scaling for on-demand instances

     Set Minimum Number of Instances to 0 and Maximum Number of Instances to the maximum number of on-demand instances. If you do not configure Maximum Number of Instances, the limit defaults to the maximum instance limit for your account in the current region.

     Note

     Idle Mode, Scheduled Setting Modification, and Metric-based Setting Modification take effect only in provisioned mode.

   • Configure scaling for provisioned instances

     Parameter	Description
     Basic Settings
     Version or Alias	Select the version or alias for which you want to create provisioned instances. Note Provisioned instances can be created only for the LATEST version.
     Minimum Number of Instances	Enter the number of provisioned instances. Minimum Number of Instances = Number of provisioned instances. Note By limiting the minimum number of function-level instances, you can ensure quick responses to invocations, reduce cold starts, and better serve latency-sensitive online businesses.
     Idle mode	Select whether to enable or disable idle mode. It is disabled by default. The following describes the options: If you enable this feature, vCPUs are allocated to provisioned instances only while they are processing requests. At other times, the instance's CPU is frozen. When idle mode is enabled, Function Compute prioritizes routing requests to the same instance based on the function's instance concurrency until that instance is at full capacity. For example, assume a function has an instance concurrency of 50 and you have 10 idle provisioned instances. If Function Compute receives 40 requests simultaneously, all 40 requests are routed to a single instance, which then transitions from idle to active. If you disable this feature, provisioned instances are allocated vCPUs regardless of whether they are processing requests.
     Maximum Number of Instances	Enter the maximum number of instances. Maximum Number of Instances = Number of provisioned instances + Maximum number of on-demand instances. Note By limiting the maximum number of function-level instances, you can prevent a single function from occupying too many instances due to excessive invocations. This protects backend resources and avoids unexpected costs. If you leave this parameter empty, the limit defaults to the maximum instance limit for your account in the current region.
     (Optional) Scheduled Setting Modification: You can create scheduled scaling rules to configure provisioned instances more flexibly. It sets the number of provisioned instances to a specified value at a designated time to better align with your business's concurrency demands. For more information about how it works and for configuration examples, see Scheduled scaling.
     Policy Name	Enter a custom policy name.
     Minimum Number of Instances	Set the number of provisioned instances as needed.
     Schedule expression (UTC)	The schedule information. Example: cron(0 0 20 * * *). For more information, see Parameter description.
     Effective time (UTC)	The period during which the scheduled auto scaling rule is active.
     (Optional) Metric-based Setting Modification: Scales provisioned resources once per minute based on the utilization of various instance metrics or the provisioned instance concurrency utilization. For more information about how it works and for configuration examples, see Metric-based scaling.
     Policy Name	Enter a custom policy name.
     Minimum range of instances	Set the allowable range (minimum and maximum) for the number of provisioned instances adjusted by this policy.
     Utilization type	This parameter is valid only when the function's instance type is a GPU instance. Select the metric type for the utilization-based auto scaling policy. For more information about auto scaling policies for GPU instances, see Auto scaling policies for provisioned GPU mode.
     Concurrency Usage Threshold	Set the target utilization threshold. When utilization falls below this threshold, a scale-in occurs. When utilization exceeds this threshold, a scale-out occurs.
     Effective time (UTC)	The period during which the metric-based auto scaling rule is active.

Modify or delete an auto scaling rule

On the Auto Scaling page, you can view the list of created rules. You can find the target rule in the list and click Modify or Delete in the Actions column.

Auto scaling methods for provisioned mode

To avoid the low utilization that can result from a fixed number of provisioned instances, you can use scheduled and metric-based scaling.

Scheduled scaling

• Definition: Scheduled scaling sets the number of provisioned instances to a specific value at a scheduled time, which helps align capacity with predictable concurrency demands.

• Use cases: This method is suitable for functions with obvious cyclical patterns or predictable traffic peaks. When the function invocation concurrency is greater than the scheduled provisioned value, the excess load is handled by on-demand mode instances.

• Configuration example: The following figure shows a configuration with two scheduled actions. Before the function invocation traffic arrives, the first scheduled configuration scales out the provisioned instances to a larger value. After the traffic decreases, the second scheduled configuration scales in the provisioned instances to a smaller value.

{
  "ServiceName": "service_1",
  "FunctionName": "function_1",
  "Qualifier": "alias_1",
  "ScheduledActions": [
    {
      "Name": "action_1",
      "StartTime": "2022-11-01T10:00:00Z",
      "EndTime": "2022-11-30T10:00:00Z",
      "TargetValue": 50,
      "ScheduleExpression": "cron(0 0 20 * * *)"
    },
    {
      "Name": "action_2",
      "StartTime": "2022-11-01T10:00:00Z",
      "EndTime": "2022-11-30T10:00:00Z",
      "TargetValue": 10,
      "ScheduleExpression": "cron(0 0 22 * * *)"
    }
  ]
}

Metric-based scaling

• Definition: Dynamically scales function instances in provisioned mode by tracking monitoring metrics.

• Use case: The Function Compute system periodically collects metrics on the concurrency or resource utilization of provisioned instances. The system uses these metrics, along with your configured scale-out and scale-in trigger values, to control the scaling of function instances in provisioned mode. This process ensures that the number of provisioned instances closely aligns with your actual resource usage.

• How it works: Metric-based tracking auto scaling adjusts provisioned resources once per minute based on metric conditions.

  • When the metric exceeds the scale-out threshold, an aggressive policy begins to scale out the number of provisioned instances, quickly expanding to the target value.

  • When the metric falls below the scale-in threshold, a conservative policy begins to scale in the number of provisioned instances, gradually approaching the scale-in target.

  If you have set maximum and minimum scaling values in the system, the number of provisioned function instances will scale between these values. Scaling out stops when the maximum value is reached, and scaling in stops when the minimum value is reached.

• Configuration example: The following figure shows an example of scaling based on the provisioned instance utilization metric.

  • As traffic increases, the scale-out threshold is triggered, and the provisioned function instances begin to scale out. When the configured maximum value is reached, scaling out stops, and any excess requests are allocated to on-demand function instances.

  • As traffic decreases, the scale-in threshold is triggered, and the provisioned function instances begin to scale in.

  

The concurrency utilization of provisioned function instances is based only on the concurrency of provisioned instances and does not include data from on-demand mode.

Metric calculation: The ratio of the number of concurrent requests being handled by provisioned function instances to the maximum number of concurrent requests that all provisioned function instances can handle. The value ranges from 0 to 1.

{
  "ServiceName": "service_1",
  "FunctionName": "function_1",
  "Qualifier": "alias_1",
  "TargetTrackingPolicies": [
    {
      "Name": "action_1",
      "StartTime": "2022-11-01T10:00:00Z",
      "EndTime": "2022-11-30T10:00:00Z",
      "MetricType": "ProvisionedConcurrencyUtilization",
      "MetricTarget": 0.6,
      "MinCapacity": 10,
      "MaxCapacity": 100,
    }
  ]
}

Related documents

• For basic concepts and billing methods of on-demand and provisioned modes: Instance types and usage modes.

• To view the number of provisioned instances in use after you configure auto scaling, check the FunctionProvisionedCurrentInstance metric: Function-level metrics.