Help Center/ Cloud Container Engine/ User Guide/ Auto Scaling/ Node Scaling/ Modifying the Node Scale-In Concurrency Settings
Updated on 2025-11-06 GMT+08:00

Modifying the Node Scale-In Concurrency Settings

After a node auto scaling policy is configured, scale-in latency may become excessive or scale-in speed may fall short of expectations, particularly in large-scale clusters. To address these issues, the CCE Cluster Autoscaler add-on provides configurable parameters that allow you to set the maximum garbage collection (GC) waiting time for individual nodes and define the number of nodes that can be scaled in concurrently. These settings can be flexibly adjusted based on specific service requirements to improve scaling efficiency.

By setting maxGracefulTerminationFlag, you can specify a graceful termination period of up to 60 minutes. This ensures that service pods have enough time to complete ongoing tasks before being terminated. Additionally, maxScaleDownParallelism, maxDrainParallelism, and parallelDrain allow you to control the number of concurrent node scale-in operations. Fine-tuning these settings helps accelerate the scale-in process and improve overall efficiency.

Prerequisites

The CCE Cluster Autoscaler add-on of 1.27.182 or later has been installed in the cluster.

Procedure

  1. Log in to the CCE console and click the cluster name to access the cluster console.
  2. In the navigation pane, choose Add-ons. Locate CCE Cluster Autoscaler on the right and click Edit. If the add-on is not installed in the cluster, install it first.
  3. In the sliding window, click Edit YAML above the add-on name.
  4. Find spec.values.custom and modify the parameters below. For details about related parameters, see CCE Cluster Autoscaler Parameters.

    • parallelDrain: whether to enable parallel drainage of nodes
    • maxScaleDownParallelism: the maximum number of concurrent scale-in tasks for both empty and non-empty nodes
    • maxDrainParallelism: the maximum number of nodes that can be drained in parallel
    • maxGracefulTerminationFlag: the maximum waiting time for the graceful termination of a node (in seconds)

  5. Click Submit.

Verifying the Modification

You can view the logs of the add-on pods to ensure that the new startup parameters (--parallel-drain, --max-scale-down-parallelism, --max-drain-parallelism, and --max-graceful-termination-sec) have been added.

kubectl logs -f -n kube-system -l app=cluster-autoscaler

CCE Cluster Autoscaler Parameters

Add-on Parameter

Community Parameter

Description

Example Value

parallelDrain

parallel-drain

  • Whether to enable parallel drainage of nodes.
  • If it is set to true, the add-on will drain multiple nodes in parallel to accelerate the scale-in process. It defaults to false. This specifies that parallel drainage is disabled.

true

maxScaleDownParallelism

max-scale-down-parallelism

  • The maximum number of concurrent scale-in tasks for both empty and non-empty nodes.
  • It controls the maximum number of concurrent scale-in tasks for both empty and non-empty nodes with --parallel-drain set to true. If it is set to 20, it indicates that up to 20 nodes will be moved in parallel at a time. It defaults to 10.

20

maxDrainParallelism

max-drain-parallelism

  • The maximum number of nodes that can be drained in parallel.
  • It controls the maximum number of nodes for parallel drainage with --parallel-drain set to true. If it is set to 20, it indicates that up to 20 nodes will be drained in parallel at a time. It defaults to 2, and the upper limit is the same as the value of max-scale-down-parallelism.

20

maxGracefulTerminationFlag

max-graceful-termination-sec

  • The maximum waiting time (in seconds) for the graceful termination of a node.
  • If a node is marked as terminated, the add-on will wait for a period so that pods on the node can be gracefully terminated. This parameter controls the maximum waiting time. If it is set to 3600, the add-on allows up to 1 hour for termination. It defaults to 600 (10 minutes).

3600