Setting Cluster Auto Scaling

Scenario

The Cluster Auto Scaling feature allows CCE to automatically scale out a cluster (adding worker nodes to a cluster) according to custom policies when workloads cannot be scheduled into the cluster due to insufficient cluster resources.

Notes and Constraints

• Currently, master nodes cannot be automatically added to or removed from clusters.
• If both auto scale-in and auto scale-out are required, use the autoscaler add-on. For details, see autoscaler.
• Clusters of v1.17 do not support auto scaling using AOM. You can use node pools for auto scaling. For details, see Node Pool Overview.

Automatic Cluster Scale-out

1. Log in to the CCE console. Choose Resource Management > Clusters in the navigation pane. In the card view of the cluster to be scaled, choose More > Auto Scaling.
2. Click the Scale-out Settings tab and then Edit. Set the maximum number of nodes, minimum number of nodes, cooldown period, and node configuration.

   Table 1 Scale-out settings

   Parameter	Description
   Cooldown Period	Interval between consecutive scale-out operations, in the unit of second. The cooldown period ensures that a scale-out operation is initiated only when previous scaling operation is finished and the system is running stably. The value ranges from 60 to 3600, in seconds. The default value is 900. If the cooling interval is less than 900 seconds (15 minutes), the auto scaling may not work well, because creating a node may take 2 to 10 minutes.
   Maximum Nodes	Maximum number of nodes to which the cluster can scale out. 1 ≤ Maximum Nodes < cluster node quota NOTE: The cluster node quota depends on the cluster size (maximum number of nodes that can be managed by a cluster) and the node quota of the account. The cluster node quota used here is the smaller of the two.
   Node Configuration	If scale-out is required after the scale-out policy is executed, the system creates a node. Click Set and set the node parameters. For details about how to set the node parameters, see Creating a Node. After the parameters are configured, click Submit.

3. After confirming the scale-out configuration and node parameters, click OK.
4. Set the scale-out policy for the cluster. Click the Scale-out Policies tab and click Add Policy.
   • Policy Name: Enter a policy name, for example, policy01.
   • Policy Type: Currently, the following types of auto scale-out policies are supported:
     • Metric-based policy: Scale-out is performed based on the CPU or memory settings.

       Table 2 Parameters for adding a metric-based policy

       Parameter	Description
       *Metric	Select Allocated CPU or Allocated Memory.
       *Trigger Condition	Set a condition for triggering a scale-out policy, that is, when the average CPU or memory allocation value is greater than or less than a specified percentage.
       *Monitoring Window	Size of the data aggregation window. Select a value from the drop-down list. If you select 15min, the selected metric is measured every 15 minutes.
       *Threshold Crossings	Number of consecutive times that the threshold is reached within the monitoring window. The calculation cycle is fixed at one minute. If you set this parameter to 3, the configured action will be triggered when the metrics meet the specified threshold for three consecutive times.
       *Action	Action executed after a policy is triggered.

     • Scheduled policy: Scale-out is performed at a specified time.

       Table 3 Parameters for adding a scheduled policy

       Parameter	Description
       *Policy Type	Set this parameter to Scheduled policy.
       *Trigger Time	Time at which the policy is triggered.
       *Action	Action executed after a policy is triggered.

     • Periodic policy: Scale-out can be performed by day, week, or month.

       Table 4 Parameters for adding a periodic policy

       Parameter	Description
       *Policy Type	Set the parameter to Periodic policy.
       *Time Range	Specify the time for triggering the policy.
       *Action	Action executed after a policy is triggered.

5. Click OK.

   After the auto scale-out is completed, choose Resource Management > Nodes in the navigation pane. On the node list, you can view the worker nodes added during cluster auto scaling.