ACK clusters - Container Service for Kubernetes - Alibaba Cloud Documentation Center

Container Service for Kubernetes offers multiple cluster types. These clusters come with different features, O&M requirements, and service level agreements (SLAs), making them suitable for various scenarios. This topic compares cluster types to help you select the one that best suits your business needs.

Cluster types

ACK provides two types of clusters, which are distinguished by whether Alibaba Cloud manages the control plane:

ACK managed clusters: The control plane of an ACK managed cluster is fully managed by Alibaba Cloud. They are available in Pro and Basic editions, which differ in control plane availability and advanced customization features. The Pro edition supports the reserved control plane feature (Pro XL / Pro 2XL / Pro 4XL), which allows you to pre-allocate and reserve control plane resources to ensure predictable control plane performance.
ACK dedicated clusters: You must create and maintain the control plane of an ACK dedicated cluster.
Important
ACK dedicated clusters can no longer be created. For more information, see Product Announcement: Discontinuation of New ACK Dedicated Clusters.

The following table describes the differences between the cluster types.

Item	ACK managed cluster		ACK dedicated cluster
	Pro	Basic
Cluster scale	Up to 100 clusters per account. By default, each cluster supports up to 5,000 worker nodes. You can request a quota increase in Quota Center.	Up to 2 clusters per account. By default, each cluster supports up to 10 worker nodes. Quota increases are not supported.	Up to 100 clusters per account. By default, each cluster supports up to 5,000 worker nodes. You can request a quota increase in Quota Center.
Management scope	Supports auto mode: Enabled: You only need to perform simple planning and configuration to create the cluster. The cluster control plane and key components are fully managed, and a node pool with auto mode enabled is created by default. Disabled: The cluster control plane is fully managed, and you are responsible for maintaining the worker nodes. Supports reserved control plane: By reserving control plane resources and API Server baseline configurations, this feature provides predictable performance by design and eliminates the uncertainty of reactive scaling. You can upgrade or downgrade between Pro, Pro XL, Pro 2XL, and Pro 4XL editions.	The cluster control plane is fully managed. You are responsible for maintaining the worker nodes.	The cluster control plane is not managed. You are responsible for maintaining both the master and worker nodes.
Scenarios	Enterprise production and testing environments. Scenarios where cost reduction is a priority. Scenarios where you want to focus on business applications and reduce investment in cluster O&M.	Suitable for personal learning and testing due to its small cluster scale.	Ideal for users with the Kubernetes expertise to independently plan, manage, and maintain clusters, especially when cost is not a primary concern. Scenarios that require research and deep customization of Kubernetes, such as custom requirements for the cluster control plane (master nodes).
Billing	You are charged cluster management fees based on the number of clusters. You are also charged for other Alibaba Cloud services used by worker nodes and some components, such as Simple Log Service (SLS).	Cluster management is free of charge. You are charged for other Alibaba Cloud services used by worker nodes and some components, such as Simple Log Service (SLS).	Cluster management is free of charge. You are charged for master nodes, worker nodes, and other Alibaba Cloud services used by some components, such as Simple Log Service (SLS).
SLA	A service availability of 99.95% is guaranteed for regional clusters, and 99.50% for zonal clusters. For more information, see Container Service for Kubernetes Service Level Agreement.	No SLA is provided.

ACK managed Pro cluster advantages

The following table compares the capabilities of ACK managed Pro clusters and ACK managed Basic clusters.

Note

In the following table, indicates that a feature is supported, and indicates that a feature is not supported.

Feature	ACK managed Pro	ACK managed Basic
Custom parameter settings for control plane components
API Server monitoring metrics
High-frequency hot and cold backups with geo-disaster recovery for etcd
etcd observability monitoring metrics
Gang scheduling policy
CPU topology-aware scheduling
GPU topology-aware scheduling
GPU sharing with fine-grained scheduling
At-rest encryption for Secrets by using KMS
Managed node pool

Hot migration

Both ACK managed Basic clusters and ACK dedicated clusters support hot migration to ACK managed Pro clusters. For more information, see the following topics:

Auto mode

When you create an ACK managed cluster, you can enable auto mode. After you enable this mode, you can quickly create a Kubernetes cluster that complies with best practices, requiring only simple network planning. Auto mode provides the following features:

Fully managed O&M: The cluster control plane and key components are fully managed. By default, a node pool with auto mode enabled, known as an auto mode node pool, is created. This node pool dynamically scales in or out based on workload demands, and ACK handles O&M tasks such as OS upgrades and security patching.
Intelligent resource provisioning: Optimal instance types are automatically recommended, eliminating the need for manual configuration.
Optimized base software stack: The immutable root file system of ContainerOS enhances security. A streamlined system and configuration accelerate node startup, while an optimized kernel maximizes hardware performance.

Use auto mode in the following scenarios:

Dynamic and elastic resource scheduling: In scenarios where workload demands fluctuate significantly, auto mode can quickly respond to changes by automatically scaling computing resources, which reduces cluster resource costs.
DevOps and CI/CD pipelines: In continuous integration and deployment (CI/CD) environments, auto mode can automatically adjust resources based on build and testing requirements to improve development efficiency and reduce costs.

Auto mode is designed based on the concepts of elastic capacity, immutable infrastructure, and zero-O&M. If your business heavily relies on node environment customization or node-local persistent storage, perform a comprehensive application assessment to identify potential compatibility risks before migration.

Important

Auto mode is designed to provide automated and intelligent O&M for Kubernetes clusters. However, you still have some responsibilities in certain scenarios. For more information, see Shared responsibility model.

Features

Feature	Description
Cluster management	Cluster creation: You can create various types of clusters, choose from a wide range of worker nodes, and apply flexible custom configurations. For more information, see Create an ACK managed cluster and Create an ACK dedicated cluster (Discontinued). Cluster upgrades: You can automatically or manually upgrade the Kubernetes version of a cluster and manage component upgrades from a single place. For more information, see Manually upgrade a cluster and Automatically upgrade a cluster. Auto scaling: You can use one-click vertical scaling to quickly respond to business fluctuations. Service-level affinity policies and horizontal scaling are also supported. Scheduling: ACK supports hybrid scheduling of different elastic resources, fine-grained scheduling of heterogeneous resources, and task scheduling for batch computing to improve application performance and overall cluster resource utilization. Multi-cluster management: You can connect clusters from on-premises data centers and multiple clouds across different regions for unified hybrid cloud application management. Authorization management: Supports RAM authorization and RBAC permission management.
Nodes and node pools	You can manage the lifecycle of node pools and configure node pools with different specifications in the same cluster, such as vSwitches, container runtimes, OSes, and security groups. For more information, see Nodes and Node pools.
Application management	Application creation: You can create various types of applications from images and templates, and configure settings such as environment variables, application health checks, data disks, and logs. Full-lifecycle management: Supports viewing, updating, and deleting applications, rolling back applications to previous versions, viewing application events, performing rolling updates, replacing applications, and redeploying applications by using triggers. Application scheduling: Supports three scheduling policies: node affinity, pod affinity, and pod anti-affinity. Application scaling: Supports manual scaling of application container instances and HPA auto scaling policies. Application release: Supports canary releases and blue-green deployments. Includes an application catalog that simplifies cloud service integration. App Center: After an application is deployed, App Center provides a unified view of the application topology and manages versions and rollbacks for scenarios such as continuous deployment. Application backup and recovery: Supports backing up and restoring Kubernetes applications. For more information, see Back up and restore applications in a cluster.
Storage	Storage plug-ins: Supports CSI storage plug-ins. For more information, see Storage. Volumes and persistent volume claims: You can create block storage, NAS, and OSS volumes. Supports mounting volumes by using persistent volume claims (PVCs). Supports dynamic creation and migration of volumes. Supports viewing and updating volumes and persistent volume claims by using scripts.
Network	Supports two container network plug-ins: Terway and Flannel. For more information, see Networking. Supports defining CIDR blocks for Services and Pods. Supports NetworkPolicy. For more information, see Use network policies in ACK clusters. Supports Ingress. For more information, see Ingress management. Supports service discovery by using DNS. For more information, see Service discovery and DNS.
Auto scaling	ACK automatically adjusts elastic computing resources based on business requirements and policies. This includes: Workload scaling (scheduling-layer elasticity): Modifies the scheduling capacity for workloads. Node scaling (resource-layer elasticity): Scales out node resources to provide more capacity when the cluster's current capacity cannot meet scheduling demands. For more information, see Auto scaling.
Scheduling	ACK provides various scheduling policies for different workloads, such as task scheduling, QoS-aware scheduling, and rescheduling, to improve application performance and overall cluster resource utilization. For more information, see Scheduling.
O&M and security	Observability: Monitoring: Supports monitoring at the cluster, node, application, and container instance levels. The Prometheus component is supported. Logging: Supports viewing cluster logs, collecting application logs, and viewing container instance logs. Alerting: Supports alerts for abnormal events in ACK and for container-related metrics. For more information, see Manage alerts for ACK. Cluster inspection and diagnostics (AIOps) Cluster check: You can perform a cluster check before operations such as cluster upgrades or migrations to confirm whether the cluster meets the requirements. Cluster inspection: Scans the running status of the cluster to identify potential risks, such as remaining cloud resource quotas and key Kubernetes resource levels. You can then troubleshoot risk items and fix issues based on recommended solutions. Cluster diagnostics: Provides one-click fault diagnosis capabilities, including node, pod, Service, Ingress, memory, and network diagnostics, to help you locate problems in the cluster. Cost analysis: Visualizes cluster resource usage and cost distribution to improve cluster resource utilization. Security Center: Supports runtime security policy management, application security configuration inspection, and runtime security monitoring and alerting to enhance the overall defense-in-depth capability for container security. Sandboxed-Container: Run applications in a lightweight virtual machine sandbox environment with an independent kernel, providing better security isolation. This feature is suitable for scenarios such as untrusted application isolation, fault isolation, performance isolation, and multi-tenant workload isolation. Confidential Computing: A one-stop cloud-native confidential computing platform based on Intel SGX for delivering and managing confidential applications. Protects the security, integrity, and confidentiality of data while it is in use. With confidential computing, place sensitive data and code in a special trusted execution environment (TEE).
Heterogeneous computing	GPU: You can create clusters with GPU instances as worker nodes and use features such as GPU scheduling, GPU monitoring, GPU auto scaling, and GPU O&M. For more information, see Add GPU nodes to a cluster. GPU sharing: You can use a GPU sharing and scheduling framework to run multiple containers on the same GPU device in clusters on the cloud or in your data centers. For more information, see GPU sharing. Cloud-native AI: ACK provides cloud-native AI capabilities to support the orchestration and management of data computing tasks. For more information, see Overview of the cloud-native AI suite.
Developer tools	API CLI SDK Terraform