Docs

Volcengine Kubernetes Engine

Creating clusters

Preset control plane

The preset control plane is a fixed-specification control plane capability provided by VKE for large-scale, high-concurrency, and performance-sensitive workloads. It prepares deterministic API concurrency handling capacity and Pod scheduling capacity before business peak hours. This topic describes the basic concepts, applicable scenarios, specifications, and usage of the preset control plane.

What are the preset control plane specifications

Basic concepts

When you create a VKE cluster, the system provides a professional control plane by default. This type of control plane dynamically adjusts resources based on actual load and automatically scales in or out when business load changes. This meets the usage requirements of most regular business scenarios. However, in ultra-large-scale clusters or burst or high-concurrency scenarios, it takes time for the control plane to detect load changes and complete resource scale-out. During this period, a large number of API requests or Pod scheduling tasks may experience brief waiting. This makes it difficult to meet business requirements for stability and predictability in control plane processing capacity.

To address the requirements for quick response in ultra-large-scale clusters or burst or high-concurrency scenarios, VKE provides preset control plane specifications. The preset control plane pre-allocates and fixes control plane resources based on the selected specification. This provides deterministic processing capacity for the API Server, ETCD, and Scheduler and keeps it stable. This helps you complete capacity planning in advance and reduces the impact of temporary scale-out waits on the business. The preset control plane also supports manual scale-up, manual scale-down, and automatic scale-up. When business load exceeds the processing capacity of the current specification and scale-up conditions are met, the system can perform automatic scale-out to reduce the impact of insufficient control plane capacity on business stability.

Applicable scenarios

AI training and inference: Large-scale training or inference tasks require batch creation and scheduling of Pods. The preset control plane can prepare scheduling capabilities in advance, reducing the impact of temporary control plane scale-out on task start efficiency.

Large-scale cluster: Businesses such as Agent applications and microservices continuously generate a large number of API requests and require stable and deterministic control plane performance limits.

Critical business: Businesses are sensitive to control plane response latency and need to maintain predictable API concurrency handling capacity and Pod scheduling capacity during traffic peaks.

Specification description

Specification comparison

The preset control plane provides you with multiple types of specifications and performance metrics. The following table describes the main differences and relationships between the preset and the professional control plane.

tip

[Beta - apply for trial]: The 4XL specification is currently in beta. To use it, submit an application.

Item	Professional Edition	Preset control plane (XL)	Preset control plane (2XL)	Preset control plane (4XL)
Concurrent API requests (Seats)	Dynamic. The upper limit does not exceed the specifications of the preset control plane (XL).	3,900	7,800	15,600
Pod scheduling rate (Pods/s)	Dynamic. The upper limit does not exceed the specifications of the preset control plane (XL).	200	400	600
ETCD database capacity	8 GB	16 GB	16 GB	16 GB
Control plane resource configuration	Load-based dynamic scaling	Pre-allocated fixed capacity	Pre-allocated fixed capacity	Pre-allocated fixed capacity
Kubernetes version	No version limit	Only Kubernetes v1.32 and later are supported. tip Existing clusters with a Kubernetes version earlier than v1.32 need to first upgrade Kubernetes to v1.32 or later. For details, see Upgrading clusters.
Specification switch	Scale-up to the preset control plane specification is supported. Scale-down from the preset control plane specification is supported.	Manual scale-up/down is supported. Automatic scale-up is supported, but automatic scale-down is not supported.
SLA	99.95%	99.95%	99.95%	99.95%
Applicable scenarios	Suitable for regular business that can accept brief capacity adjustment time when the control plane automatically scales based on load.	Applicable to businesses that are large-scale and high-concurrency and require stable and predictable control plane processing capability.

Billing methods

The specifications of the preset control plane are billed in the same way as those of the professional control plane. They support only the pay-as-you-go billing method.

Billing is reported on the hour. At each hour, the system charges based on the control plane specification used by the cluster.

After a specification switch, the new billing standard takes effect from the hour of the switch time. For example, if you switch from the professional control plane to the preset control plane (XL) at 09:10, the entire hourly billing period starting at 09:00 (09:00 to 10:00) is billed at the preset control plane (XL) price.

tip

For billing details of different preset control plane specifications, see Product billing.

Procedure

Use when creating clusters

Use for existing clusters

When creating a cluster, you can proactively select the preset control plane during the creation process. Procedure:

In the left navigation pane, click Cluster. On the page that appears, click Create cluster.

In the Create cluster dialog box, go to the Custom managed clusters section and click Create now.

In the Cluster configuration step, follow the page prompt to select different preset control plane specifications for the Cluster Specifications parameters in the Basic configuration section.

(Optional) When the specification is set to the preset control plane, the Auto scale-up feature can be enabled. After this feature is enabled, when the resource utilization of control plane add-ons continuously reaches the configured scale-up threshold or OOM occurs, the system will scale up the cluster control plane step by step to the next higher specification. This reduces the risk of business throttling due to insufficient control plane capacity.

Complete other cluster configurations. For parameter details, see Creating managed clusters.

After the cluster is created, you can view the current specification in the Basic information section on the Overview tab on the cluster details page. You can also view resource utilization on the Control plane observability tab.

Specification change

Considerations

The control plane will be briefly interrupted during the specification change. We recommend stopping other operations, such as creating workloads and upgrading the cluster, during the change.

The system supports automatic scale-up but does not support automatic scale-down. You need to manually perform the scale-down if needed.

After a switch between the professional and the preset control plane, historical Kubernetes events before the switch can no longer be queried. Changing between different preset control plane specifications is not subject to this restriction.

Manual scale-up/down

In the left navigation pane, click Cluster. Click the name of the target cluster to go to the management page of this cluster.

Click Overview in the left navigation pane, and then click the Overview tab on the right page.

In the Basic information section, click Change specifications next to the Cluster specifications parameter.

Select the target specification and compare the API request concurrency, Pod scheduling rate, and ETCD database capacity before and after the change.

tip

When switching from a higher preset control plane specification to a lower one (that is, a scale-down), the system automatically checks the following conditions. Scale-down is allowed only when all conditions are met.

The current CPU usage of both API Server and ETCD is lower than 20% of the CPU limit for the target specification.

The current memory usage of both API Server and ETCD is lower than 20% of the memory limit for the target specification.

The current ETCD database size is lower than 50% of the ETCD capacity for the target specification.

When a preset control plane specification is switched back to the professional edition, in addition to meeting the preceding CPU, memory, and ETCD capacity conditions, the current ETCD database size must also be less than 8 GB. If any condition is not met, the page will prompt for the specific reason, and the scale-down request cannot be submitted. Follow the prompts to reduce the CPU and memory usage of the API Server or ETCD, and clean up ETCD before retrying.

View the billing information and risk notice, and select I have read and am aware of the following precautions.

Click Confirm, and the cluster starts changing the control plane specification.

After the change is complete, click Allocation Record in the Cluster Specifications section. You can view the status, original specification, target specification, change type, trigger method, start time, and end time.

Automatic scale-up

Automatic scale-up allows the system to continuously monitor the resource utilization of the control plane. When the resource utilization of control plane add-ons continuously reaches the configured scale-up threshold or OOM occurs, the system will scale up the cluster control plane step by step to the next higher specification. This reduces the risk of business throttling due to insufficient control plane capacity.

warning

The professional edition does not involve specifications. Therefore, it does not support automatic scale-up.

The system scales up the control plane only to the next higher specification each time (for example, 2XL to 4XL). Automatic scale-up stops after the control plane reaches the highest specification.

Auto scale-down is not supported. After the switch is disabled, the current specification remains unchanged. You need to manually perform the scale-down if needed.

During automatic scale-up, control plane add-ons are updated on a rolling basis, and API access may experience brief jitter.

To prevent continuous scale-ups caused by resource jitter, the interval between two automatic scale-ups for the same cluster is 30 minutes.

After the automatic scale-up feature is enabled, the system automatically scales up the control plane when any condition in the following table is met.

Condition category	Trigger condition
Control plane add-on load	The CPU or memory utilization of either the kube-apiserver or ETCD add-on in the control plane reaches 90% for 10 consecutive minutes.
Control plane add-on OOM	OOM restarts occur in the kube-apiserver add-on.

Specify parameters during cluster creation

Existing cluster configuration

You can select the preset control plane when creating a cluster and enable Auto scale-up. For details, see Usage method.

Capacity planning and metric monitoring

Core metrics

The following table describes the main performance metrics and their meanings for different control plane specifications.

Core metrics	Metrics
Concurrent API requests (Seats)	Seats are used to measure the number of requests that the API Server can process simultaneously. Processing overheads vary based on different requests. Therefore, they may occupy different numbers of Seats. A higher upper limit for Seats in the specification indicates more API requests that can be processed simultaneously.
Pod scheduling rate	The number of Pods that the scheduler can assign to nodes per second. This is used to assess the scheduling processing capability of the control plane.
ETCD database capacity	The capacity used to store cluster state data. The preset control plane stores Kubernetes events independently, reducing the impact of high-frequency event writes on core state data.

Metric monitoring

Before configuring or changing the cluster control plane specification, we recommend evaluating API request concurrency, Pod scheduling rate, and ETCD database size based on business peaks and reserving an appropriate margin.

In the left navigation pane, click Cluster. Click the name of the target cluster to go to the management page of this cluster.

In the left navigation pane, click Overview. On the Overview page, select the Control plane observability tab to view the cluster control plane dashboard. For details, see Control plane observability.