Executive summary
Alibaba Cloud Container Service for Kubernetes (ACK) delivers a portfolio of Kubernetes offerings that let enterprises choose the level of operational responsibility they are comfortable with. Managed clusters (Basic and Pro) provide a fully‑operated control plane, while Serverless and Edge options remove node management entirely. The discontinued dedicated‑cluster model required customers to run the control plane themselves and can no longer be created. ACK One adds a hybrid‑cloud layer for multi‑cluster fleet management, backup, and disaster‑recovery across on‑premises and cloud environments.
This article walks through the technical fundamentals, architecture, implementation workflow, security model, enterprise use cases, benefits, limitations, and best practices for building an enterprise‑grade Kubernetes platform on Alibaba Cloud. All statements are drawn from the official Alibaba Cloud documentation linked in the References section.
Introduction
Kubernetes has become the de‑facto standard for container orchestration, and many global enterprises run workloads across multiple clouds and on‑premises data centers. Alibaba Cloud’s ACK service is a CNCF platinum member offering that is Kubernetes‑conformant, meaning workloads that run on any certified cluster can be moved to ACK without code changes.
ACK presents several cluster flavors:
- Managed clusters – Alibaba Cloud provisions and operates the control plane; customers create and manage worker node pools. Two service tiers exist:
- ACK Basic – entry‑level tier for testing and small workloads.
- ACK Pro – production‑grade tier with higher reliability, a service level agreement, and additional security controls.
- Serverless clusters – applications run without purchasing or maintaining nodes; the platform automatically scales the underlying compute.
- Edge clusters – optimized for low‑latency, geographically distributed workloads.
- ACK One – a distributed cloud container platform that aggregates registered clusters (on‑prem or other clouds) into a single management plane for fleet scheduling, backup, and disaster recovery.
Dedicated clusters, where the customer operated the control plane and etcd, are no longer creatable. Existing dedicated clusters remain functional but cannot be expanded, and new platforms should adopt managed clusters, preferably the Pro tier for production workloads.
Technical fundamentals
Cluster control plane – In a managed cluster, Alibaba Cloud runs the master nodes and etcd as a fully managed service. The control plane is replicated across zones for high availability. The documented availability commitment is 99.95 % for regional clusters and 99.50 % for zonal clusters. These figures are part of the Container Service for Kubernetes service‑level agreement.
Worker nodes – Customers provision Elastic Compute Service (ECS) instances or use serverless compute. Nodes are grouped into node pools, each pool sharing instance type, zone, operating system, labels, and taints. Pools are independent; changes to one pool do not affect others.
Permissions model – Two layers of access control exist:
- Resource Access Management (RAM) – Alibaba Cloud’s identity and access management service controls which cloud resources a user can create, modify, or delete. To create a cluster, a user must have the ACK service role and any additional RAM policies required for the underlying ECS resources.
- Kubernetes RBAC – Inside the cluster, standard Kubernetes role‑based access control governs what a user can do with pods, services, and other objects. RBAC bindings are granted to RAM users or RAM roles that are mapped into the cluster.
APIs and tooling – Clusters can be created and managed via the Alibaba Cloud Console, the Alibaba Cloud CLI, the ACK REST API, SDKs, or Terraform. All of these interfaces expose the same underlying operations; no undocumented commands are introduced in this guide.
Architecture and components
flowchart TB
subgraph Cloud
A[Managed Control Plane] --> B[etcd (managed)]
A --> C[API Server]
C --> D[Scheduler]
C --> E[Controller Manager]
end
subgraph Customer
F[Worker Node Pool 1] --> G[Pods]
H[Worker Node Pool 2] --> I[Pods]
J[ACK One Fleet] --> K[Registered Cluster A]
J --> L[Registered Cluster B]
end
style Cloud fill:#e6f7ff,stroke:#0078d7
style Customer fill:#fff4e6,stroke:#d9730d
A -.-> RAM[RAM Permissions]
G -.-> RBAC[RBAC]
I -.-> RBAC
K -.-> RBAC
L -.-> RBAC
The diagram illustrates a managed control plane (operated by Alibaba Cloud) linked to customer‑owned worker node pools. RAM governs access at the cloud boundary, while RBAC governs access inside each cluster. An optional ACK One fleet can sit above multiple clusters—both managed and registered—to provide unified scheduling and backup.
Detailed implementation approach
- Plan permissions – Assign the ACK service role to the identity that will create clusters. Add any additional RAM policies needed for ECS, VPC, and security groups.
- Create a managed cluster – Use the console, CLI (
aliyun ack CreateCluster), API, or Terraform (alicloud_ack_cluster). Choose Basic for proof‑of‑concept or Pro for production. Specify region, VPC, and CIDR block.
- Configure node pools – Define at least one node pool with the desired instance type, zone, OS image, and labels. Additional pools can be added later for different workloads.
- Deploy workloads – Apply Kubernetes manifests via
kubectl. Because the control plane is managed, kubeconfig files are generated by ACK and can be downloaded from the console or retrieved via the CLI.
- Enable ACK One (optional) – Register existing on‑prem or other‑cloud clusters with ACK One. Create a fleet resource to orchestrate workloads across the registered clusters and the managed clusters.
- Set up backup and disaster recovery – Use the ACK One backup center to configure scheduled snapshots of etcd and persistent volumes across all clusters in the fleet.
- Monitor and operate – Leverage Alibaba Cloud CloudMonitor, Log Service, and Container Service for Kubernetes metrics to observe node health, pod performance, and control‑plane status.
Architecture and data‑flow
| Cluster type | Who runs the control plane | New cluster creation | Availability commitment |
|---|
| ACK Basic | Alibaba Cloud (managed) | Yes | 99.95 % (regional) / 99.50 % (zonal) |
| ACK Pro | Alibaba Cloud (managed) | Yes | 99.95 % (regional) / 99.50 % (zonal) |
| Dedicated | Customer (self‑managed) | No (discontinued) | None documented |
Security and governance
- RAM policies – Define least‑privilege access for cluster creation, node pool scaling, and network configuration. Use resource‑level constraints to limit actions to specific regions or VPCs.
- Kubernetes RBAC – Map RAM identities to Kubernetes groups. Grant namespace‑level roles for developers and cluster‑admin roles for platform operators.
- Network isolation – Place clusters in dedicated VPCs with security groups that restrict inbound traffic to required ports (e.g., 443 for the API server). Use VPC peering or Alibaba Cloud Express Connect for hybrid connectivity.
- Audit – CloudMonitor logs and Log Service capture API calls made by RAM users. Enable Kubernetes audit logging to record RBAC decisions and pod‑creation events.
- Compliance – ACK is CNCF‑conformant, satisfying many regulatory requirements for container orchestration. The Pro tier adds SLA guarantees that may be required for higher‑risk workloads.
Enterprise use cases
- Global microservice platform – Deploy a fleet of Pro clusters in multiple regions. Use ACK One to schedule traffic‑aware workloads, ensuring low latency for end users.
- Hybrid cloud data processing – Register an on‑premises Kubernetes cluster with ACK One, then run batch jobs that span on‑prem and cloud nodes, leveraging the backup center for consistent snapshots.
- Edge analytics – Deploy ACK Edge clusters in remote locations to process IoT data locally, while central management remains in a managed Pro cluster.
- Serverless API backends – Use ACK Serverless clusters for event‑driven functions that scale without node management overhead.
Benefits and limitations
Benefits
- Managed control plane reduces operational burden and provides documented availability commitments.
- Multiple cluster flavors let teams select the right level of abstraction for each workload.
- ACK One adds unified multi‑cluster management, backup, and disaster‑recovery capabilities.
- Integration with Alibaba Cloud RAM simplifies cross‑service permission management.
Limitations
- Dedicated clusters can no longer be created, eliminating the option for full control over the master nodes.
- Serverless clusters abstract node details but limit access to low‑level host configuration.
- The availability commitment applies only to the control plane; worker node uptime depends on the selected ECS instance type and zone selection.
- Some advanced Kubernetes add‑ons may still require manual installation because ACK does not bundle every open‑source component.
Operational considerations
- Scaling – Node pools can be scaled vertically (changing instance type) or horizontally (adding more instances). Scaling actions are performed via the console, CLI, or Terraform.
- Upgrades – Alibaba Cloud handles control‑plane upgrades automatically. Worker node OS and Kubernetes version upgrades must be coordinated by the customer, typically using rolling updates.
- Monitoring – Enable CloudMonitor container insights for real‑time metrics. Set alerts for CPU, memory, and pod restarts.
- Backup – Use ACK One’s backup center to schedule etcd snapshots and persistent‑volume backups. Verify restore procedures regularly.
- Cost management – Managed clusters charge a monthly control‑plane fee plus per‑node usage. Serverless clusters charge per‑second compute usage, which can reduce cost for spiky workloads.
Best practices
- Start with Basic, migrate to Pro – Prototype workloads in a Basic cluster, then promote to Pro once stability and SLA requirements are confirmed.
- Separate environments – Use distinct VPCs and clusters for development, staging, and production to enforce isolation.
- Leverage node pools – Group workloads by resource profile; keep high‑CPU workloads in a dedicated pool to avoid contention with memory‑intensive services.
- Implement least‑privilege RAM – Grant only the permissions needed for each team (e.g., developers get cluster‑admin inside a namespace, not full cluster creation rights).
- Enable RBAC and audit – Map RAM identities to Kubernetes groups, enforce namespace‑level RBAC, and enable audit logs for compliance.
- Use ACK One for hybrid scenarios – Register on‑prem clusters and use fleet scheduling to balance load and improve resilience.
- Automate with IaC – Store cluster definitions, node‑pool configurations, and RBAC policies in Terraform or Alibaba Cloud Resource Orchestration Service templates for repeatable deployments.
Analysis
The documentation makes clear that managed clusters are now the only viable path for new enterprise deployments on ACK. The availability numbers (99.95 % regional, 99.50 % zonal) are commitments, not measured uptime, and should be presented as such. Dedicated clusters, while still supported for existing customers, cannot be newly provisioned, limiting their relevance for future platform builds.
ACK One introduces a higher‑level abstraction that addresses hybrid‑cloud and disaster‑recovery needs, but it does not replace the underlying managed clusters; it sits on top of them. The separation of RAM and RBAC aligns with best practices for zero‑trust perimeter security.
What the sources do not settle
- Performance benchmarks – The sources provide no quantitative data on latency, throughput, or pod startup times for managed versus serverless clusters.
- Cost comparison – While pricing models are described, the documentation does not include side‑by‑side cost analyses for Basic, Pro, or Serverless tiers under typical workloads.
- Feature roadmap – There is no public roadmap indicating whether dedicated clusters might be revived or how future ACK One capabilities will evolve.
- Third‑party integrations – The extent of native integrations with CI/CD pipelines, service meshes, or observability stacks beyond the generic API and CLI support is not detailed.
Conclusion
Alibaba Cloud’s ACK service offers a comprehensive set of Kubernetes options that let enterprises choose the right balance between operational control and managed convenience. Managed Basic and Pro clusters provide a reliable, SLA‑backed control plane, while Serverless and Edge clusters address niche use cases. ACK One extends the platform to hybrid and multi‑cluster scenarios, enabling fleet‑wide scheduling, backup, and disaster recovery. Teams should adopt managed Pro clusters for production workloads, employ node pools for workload segmentation, enforce a dual‑layer permission model with RAM and RBAC, and automate lifecycle management with Terraform or the Alibaba Cloud CLI. By following the best‑practice recommendations outlined above, organizations can build a resilient, secure, and scalable Kubernetes platform on Alibaba Cloud.
References
These published CloudExchangers articles are further desk reading. They are not sources for the technical claims above.