Container Service for Kubernetes:Enable node autoscaling

Before you begin

To use node autoscaling effectively, read node scaling first and understand the following:

Node autoscaling: how it works and its features

Which business scenarios node autoscaling supports

Important considerations before using node autoscaling

When scaling in, subscription instances are removed from the cluster but not released. To avoid extra costs, use pay-as-you-go instances when enabling this feature.

Important notes

• Before you begin, make sure you have enabled Auto Scaling (ESS).

• See Important notes for quotas and limits on node scaling.

• Node autoscaling has known limitations with certain scheduling policies. These may cause scaling results to differ from expectations. If your workloads or components use unsupported scheduling policies, adjust them as follows:

  • Solution 1: Switch to instant node elasticity.

  • Solution 2: Deploy affected workloads or components in node pools where node scaling is disabled.

    For example, deploy the ack-node-local-dns-admission-controller component in a node pool where node scaling is disabled, and declare the following node affinity requirements in the component configuration.

    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: "k8s.aliyun.com"
            operator: "NotIn"
            values: ["true"]

• The cluster-autoscaler component requires node resources during updates or deployments. If resources are insufficient, updates or deployments may fail and cause scaling errors. Ensure sufficient node resources.

This feature includes the following steps:

1. Step 1: Enable node autoscaling for the cluster: The automatic scaling policies configured for node pools take effect only after you enable the node autoscaling feature at the cluster level.

2. Step 2: Configure an elastic node pool: Node autoscaling applies only to node pools configured for auto-scaling. So, set the scaling mode of your target node pool to Auto.

Step 1: Enable node autoscaling for the cluster

1. Log on to the Container Service Management Console . In the navigation pane on the left, click Clusters.

2. On the Clusters page, click the name of your cluster. In the navigation pane on the left, click Nodes > Node Pools.

3. On the Node Pools page, click Configure next to Node Scaling.

   

4. If this is your first time using node autoscaling, follow the on-screen prompts to enable ESS and complete authorization. Skip this step if ESS is already enabled and authorized.

   • For ACK managed clusters: Complete the AliyunCSManagedAutoScalerRole authorization.

   • For ACK dedicated clusters: Authorize both the KubernetesWorkerRole role and the AliyunCSManagedAutoScalerRolePolicy system policy. Use the links below.

     

5. On the Node Scaling Configuration page, set Node Scaling Method to Auto Scaling. Configure the scaling settings and click OK.

   You can change the node scaling method later. To switch, select instant node elasticity, then carefully read and follow the on-screen instructions.

   Setting	Description
   Node Pool Scale-Out Order Policy	Random policy: When multiple node pools are eligible for scale-out, pick one at random. Default policy: When multiple node pools are eligible for scale-out, pick the one with the least resource waste. Priority policy: When multiple node pools are eligible for scale-out, pick the one with the highest priority. Set node pool priority using the Node Pool Scale-Out Priority parameter.
   Node Pool Scale-Out Priority	Set the scale-out priority for a node pool. This setting takes effect only when Node Pool Scale-Out Order Policy is set to Priority policy. Valid values: integers from 1 to 100. Higher numbers mean higher priority. Click Add to the right of the parameter. Select a node pool with auto-scaling enabled and assign it a priority. If no node pool with auto-scaling enabled is available, skip this setting now. Set priorities after completing Step 2: Configure an elastic node pool.
   Elasticity Sensitivity	Adjust how often the system checks whether scaling is needed. Default value: 60 seconds.
   During scaling, the elasticity component triggers scale-out automatically based on scheduling status. Important ECS nodes: The elasticity component may scale in a node only when all three conditions are met: Scale-In Threshold, Scale-In Trigger Delay, and Silence Time. GPU nodes: The elasticity component may scale in a GPU node only when all three conditions are met: GPU Scale-In Threshold, Scale-In Trigger Delay, and Silence Time.
   Allow Scale-In	Whether to allow node scale-in. When disabled, scale-in settings do not take effect. Set this carefully.
   Scale-In Threshold	For node pools with node autoscaling enabled, this is the ratio of requested resources (Request) to total resource capacity per node. A node may be scaled in only when this ratio falls below the threshold — meaning CPU and memory usage on the node are both below the Scale-In Threshold.
   GPU Scale-In Threshold	The scale-in threshold for GPU instances. A GPU node may be scaled in only when this ratio falls below the threshold — meaning CPU, memory, and GPU usage on the node are all below the GPU Scale-In Threshold.
   Scale-In Trigger Delay	The time between detecting a scale-in need and actually performing scale-in. Unit: minutes. Default value: 10 minutes. Important The elasticity component may scale in a node only after both the Scale-In Threshold is met and the Scale-In Trigger Delay passes.
   Silence Time	The time after the most recent scale-out during which the elasticity component does not perform scale-in. During silence time, the elasticity component does not scale in nodes but still checks whether nodes qualify for scale-in. After silence time ends, if a node meets both the scale-in threshold and scale-in trigger delay, the elasticity component performs scale-in normally. For example, if silence time is 10 minutes and scale-in trigger delay is 5 minutes, the elasticity component will not scale in nodes for 10 minutes after the last scale-out. But within those 10 minutes, it checks whether nodes meet scale-in conditions. Once silence time ends and the node meets the scale-in threshold and has waited at least 5 minutes, the elasticity component proceeds with scale-in.

   Advanced configuration settings

   Setting	Description
   Pod Termination Timeout	The maximum time to wait for pods on a node to terminate during scale-in. Unit: seconds. If pods fail to drain before timeout, the node will not be released in this scale-in operation.
   Minimum Pod Replicas	A scale-in protection threshold for applications managed by ReplicationController or ReplicaSet. If the current number of replicas for such an application falls below this value, nodes running its pods will not be scaled in. This setting applies only to pods managed by ReplicationController or ReplicaSet. It does not apply to pods managed by StatefulSet, DaemonSet, or other controllers.
   Enable DaemonSet Pod Drain	When enabled, the elasticity component evicts DaemonSet pods during node scale-in.
   Skip Nodes with kube-system Namespace Pods	When enabled, the cluster ignores nodes running pods in the kube-system namespace during automatic node scale-in. This ensures those nodes are not scaled in. Note This setting does not apply to DaemonSet pods or Mirror pods.

Step 2: Configure an elastic node pool

You can either configure an existing node pool by changing its Scaling Mode to Auto, or create a new node pool with auto-scaling enabled.

For details, see Create and manage node pools. Key settings include the following:

Step 3: (Optional) Verify the result

After completing the steps above, you can use node autoscaling. The node pool shows that auto-scaling has started, and the cluster has automatically installed the cluster-autoscaler component.

FAQ