Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11For managed Ray on Azure Kubernetes Service (AKS), the closest product match documented by Microsoft is Anyscale on Azure. Anyscale hosts its control plane in Azure, while Ray workloads run in the customer’s AKS environment. It is marked Public Preview, has no SLA, and is available only in a limited set of regions. A separate option is to operate Ray on AKS yourself with KubeRay and Kueue; that gives you a different operating model, not the same managed service.
What “managed Ray on AKS” means
Microsoft Learn describes Anyscale on Azure as “a managed platform for running distributed Python workloads on Ray.” The platform separates management from workload execution:
- Control plane: Hosted by Anyscale in Azure. It handles scheduling, monitoring, job management, and the console.
- Data plane: Runs in the customer’s Azure subscription on AKS. Ray workloads, container images, and data remain in that environment.
Teams can access the platform through the Azure portal, Anyscale console, CLI, or SDK, subject to documented permissions and command limitations. The documented integrations include AKS for compute; Azure Blob Storage and Azure Data Lake Storage for artifacts and datasets; Azure Container Registry for custom images; and Azure Load Balancer for access to clusters and services. Managed identities govern access to cloud resources and can be shared or mapped more granularly.
This is not the same as Microsoft operating every part of the customer’s Ray environment. The managed platform provides the control plane, but its documented restrictions and AKS responsibilities still matter when choosing it.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
How Anyscale on Azure differs from KubeRay and Kueue
Microsoft documents two distinct approaches. Anyscale on Azure is a managed Ray platform deployed onto a customer’s AKS cluster. KubeRay and Kueue are open-source components a customer deploys and operates on AKS: KubeRay manages Ray cluster lifecycle, while Kueue controls workload admission against resource quotas.
| Decision point | Anyscale on Azure | Self-managed Ray with KubeRay and Kueue |
|---|---|---|
| Operating model | Anyscale hosts the control plane; workload data plane runs in the customer’s AKS environment. | Customer deploys and operates the AKS infrastructure and open-source operators. |
| Cluster and workload handling | The managed platform handles scheduling, monitoring, and job management, subject to preview limitations. | KubeRay manages Ray cluster lifecycle; Kueue admits workloads against configured quotas. |
| Deployment approach | Provision and manage clouds through the Azure portal where required; use documented console, CLI, or SDK capabilities. | Microsoft’s reference architecture uses Terraform, Helm-installed operators, and Kubernetes manifests. |
| Availability and support posture | Public Preview, no SLA, and limited regional availability in the Microsoft Learn overview. | AKS support does not extend its SLA, limited warranty, or Azure support to the open-source components named in Microsoft’s samples. |
| Costs | Current pricing not established in the cited overview. | A complete cost comparison is not stated in the cited guidance. |
The choice is therefore less about whether Ray can run on AKS—it can in either model—and more about who operates its management layer, which features are available, and what support boundaries your team can accept.
What to check before choosing Anyscale on Azure
Microsoft’s overview identifies significant Public Preview limitations. Verify current service terms, tenant eligibility, region support, and feature status before committing a production design; availability and preview details can change.
- Deployment type: Only AKS-based deployments are supported in the overview. VM stack features and Anyscale-hosted clouds are unavailable.
- Cloud lifecycle: Creating and deleting clouds requires the Azure portal. Several CLI commands are unsupported.
- Scheduling: Workload priority applies to jobs and workspaces, not services. Job queues are unsupported.
- Other unavailable features: The overview lists unsupported machine pools, Global Resource Scheduler, lineage tracking, and selected console organization settings for billing, budgets, resource notifications, and cost analysis.
- Multiple cloud resources: A service can attach multiple cloud resources, but a Ray cluster remains within one resource; a workload is not autoscaled or scheduled across resources. Jobs can use multiple resources with fallback in a specified failure-to-start scenario. Workspaces use one resource without fallback, and services use only the primary resource.
The overview does not establish a complete supported-region list or tenant-specific access eligibility. Check the current supported-region information and your own tenant’s access during planning rather than assuming a region or subscription is eligible.
Rank #3
How the self-managed KubeRay and Kueue pattern works
Microsoft’s AKS reference architecture combines infrastructure provisioning with Kubernetes-level workload admission. Terraform provisions AKS, GPU node pools, Blob Storage, workload identity, and the Helm-installed KubeRay and Kueue operators. Kubernetes manifests then define resource types and queue policies.
- Provision the platform: Create the AKS cluster and supporting Azure resources with the Terraform configuration. The default GPU example uses a
Standard_ND96amsr_A100_v4VM node with 8 × A100 80 GB GPUs. - Define available resources and quotas: Configure
ResourceFlavorsfor CPU/GPU types,ClusterQueuesfor quotas and admission rules, and namespace-scopedLocalQueuesas workload submission points. - Submit a Ray workload: Use a RayJob or RayService Kubernetes resource. In the documented job flow, the Ray workload starts suspended.
- Admit and run it: Kueue checks quota and unsuspends admitted workloads; KubeRay then creates the Ray cluster and runs the job.
The guide’s examples cover weather-model fine-tuning, LLM training, batch inference, and online serving. The sample VM is an infrastructure example, not a general Ray minimum, performance guarantee, or price estimate. Regional GPU quota must be available for the default GPU configuration. The guide says the sample can be deployed with GPUs disabled to validate infrastructure and queues, but workloads requesting GPUs will remain Pending.
Prerequisites in Microsoft’s guide
The infrastructure guide lists Azure CLI 2.70 or later, Terraform 1.6 or later, kubectl 1.28 or later, and Python 3.10 or later for Aurora data generation. An Azure subscription is also required; the default GPU configuration additionally requires regional GPU quota. These versions are time-sensitive—Microsoft’s AKS Ray overview and infrastructure guide were last updated July 7, 2026, so recheck the current guide before deployment.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What AKS still leaves to your platform team
AKS is managed at the Kubernetes control-plane level; that does not remove responsibility for the workload environment. Microsoft’s AKS architectural guidance says the platform team configures node pools, scaling, and networking and monitors cluster infrastructure. Its support policy assigns customers responsibility for application deployments and images, identities and access, workload monitoring, and disaster recovery.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Best Value
Upgrade and scaling responsibilities are shared in specific ways: Microsoft provides supported Kubernetes versions and deprecation timelines, while customers trigger and schedule upgrades. Microsoft supplies updated node images, while customers select an auto-upgrade channel or apply updates. Customers set worker-node scaling policies, minimum and maximum values, and priorities. Plan quotas, upgrades, observability, backups, and recovery as part of the platform design rather than assuming a managed control plane covers them.
For KubeRay and Kueue, there is an additional support boundary: Microsoft’s AKS Ray/Kueue documentation says open-source software referenced in its documentation and samples is excluded from AKS service-level agreements, limited warranty, and Azure support. Teams using this pattern should establish support with the relevant projects or maintainers, or provide that operational expertise themselves.
Choose the operating model that fits your team
Choose Anyscale on Azure when
- You want a managed Ray control plane while keeping the workload data plane in your Azure subscription on AKS.
- The current preview status, lack of SLA, region limitations, and feature restrictions fit your use case and risk tolerance.
- Your required workflows do not depend on unavailable features such as job queues, lineage tracking, or cross-resource cluster scheduling.
Choose KubeRay and Kueue when
- You want to assemble and operate the Ray-on-AKS management pattern using Kubernetes resources and quota-based admission.
- Your team can own the AKS infrastructure, operator lifecycle, workload operations, and open-source support boundary.
- You can confirm that the needed regional GPU capacity and quota are available if your workloads require GPUs.
Microsoft’s documentation does not provide a complete cost comparison between these options, and current Anyscale pricing is not established there. Compare the operating responsibilities and support posture directly; do not infer that either route is cheaper from the architecture descriptions alone.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
