Docs

Volcengine Kubernetes Engine

Scheduling management

Configuring custom scheduling parameters

Volcengine Kubernetes Engine (VKE) supports custom parameters for the scheduler-plugin add-on. This allows you to specify custom scheduler behavior to meet your pod scheduling requirements. This topic describes how to configure custom scheduler parameters.

Prerequisites

The scheduler-plugin add-on has been installed. For detailed operations, see scheduler-plugin.

Limit

Only clusters that use the following Kubernetes versions support custom scheduler parameters:

Kubernetes V1.26: v1.26.6-vke.7 or later.

Kubernetes V1.24: v1.24.15-vke.18 or later.

Kubernetes V1.20: v1.20.15-vke.23 or later.

Other later Kubernetes versions released by VKE.

If your Kubernetes version does not meet the requirements, upgrade the control plane of your cluster. For details, see Upgrading clusters.

Resource weights in Binpack and Spread policies

After you configure a node scheduling policy, you can specify custom resource weights for the Binpack and Spread scheduling policies.

In the left navigation bar, click Cluster. On the page that appears, click the name of the cluster that you need to configure.

On the cluster management page, click Addons in the left navigation bar.

On the Scheduling tab, find the scheduler-plugin and click Configuration.

In the Node scheduling policy section, modify the Resource weight configuration as needed.

tip

If no custom resource weight is configured, the Binpack/Spread policy uses the default weight 1 for CPU and memory resources.

Parameters	Description
Resources	The following resource types are supported: CPU Memory GPU (evaluated based on the computing power of NVIDIA-type GPU resources) mGPU (evaluated based on the computing power of mGPU resources) GPU (NPU-A2) tip VKE follows the resource weight policy of native Kubernetes. For details, see RequestedToCapacityRatio. When you configure GPU and mGPU resource weights, you can combine node scheduling with GPU card scheduling. For details, see Combinations of Binpack and Spread scheduling policies.
Weight	The weight of a selected resource ranges from 1 to 10. A higher weight produces a higher score. That means when the system schedules pods to a node, the available capacity of the target resource on the node is prioritized.

Combinations of Binpack and Spread scheduling policies

If your workload involves GPU compute nodes and GPU cards, you can combine the two scheduling policies as described below:

tip

Node-level Binpack and Spread scheduling policies are effective for both mGPU and regular GPUs.

Node scheduling policy	GPU card scheduling policy	Scheduling mode	Application scenario
Spread (default)	Binpack (default)	Pods are spread to nodes with low GPU allocation rate. Containers are preferentially scheduled to GPUs with high packing rates on nodes.	High fault isolation performance is achieved among nodes and a small number of fragmented resources exist.
Spread	Spread	Pods are spread to nodes with low GPU allocation rate. Containers are prioritized for GPU cards with low packing rates on the node.	Nodes and GPU cards have strong fault isolation, but fragmented resources are relatively abundant. It is applicable to high-availability scenarios.
Binpack	Binpack	Pods are prioritized for nodes with high GPU allocation rates. Containers are preferentially scheduled to GPUs with high packing rates on the node.	Fault isolation is relatively weak, but resource fragmentation is minimal. This prevents resource waste caused by resource fragmentation from shared services. It is applicable to GPU high-utilization scenarios.
Binpack	Spread	Pods are preferentially scheduled to nodes with high GPU allocation rates. Containers are prioritized for GPU cards with low packing rates on the node.	GPU cards have strong fault isolation, and a large number of fragmented resources exist.

Last updated: 2026.08.03 16:44:10