Fleets

Fleets automatically add or remove nodes in your Cloudfleet Kubernetes Engine (CFKE) cluster as your workload demands change. CFKE lets you deploy, run, and monitor applications in your own cloud provider account using a unified Kubernetes cluster. You can use your existing discounts and credits and avoid vendor lock-in. CFKE automatically selects optimal instance types and runs workloads in your account. When needed, it bursts to other providers.

Fleets introduction

A Fleet defines the cloud provider where Cloudfleet provisions the infrastructure. A single Fleet contains the configuration for one or more cloud provider accounts where Cloudfleet can provision nodes for your workloads.

Every CFKE cluster can have one or more Fleets and run workloads on infrastructure from multiple Fleets at the same time. You get the best pricing and performance for your workloads and avoid vendor lock-in.

CFKE provisions infrastructure based on the workloads deployed in the Kubernetes cluster. It uses the scheduling preferences in the workload specifications to decide where resources run and how much infrastructure is needed. CFKE then provisions the required nodes through the cloud provider’s APIs. CFKE also handles the full node lifecycle, including scaling, upgrades, and deletions, so you can focus on your applications instead of infrastructure.

To learn more about node auto-provisioning, see the Node auto-provisioning documentation. For a comparison with traditional Kubernetes node pools and self-managed nodes, see How Cloudfleet provisions nodes.

Supported providers

Node auto-provisioning with Fleets is available for selected Cloud Service Providers (CSPs) that integrate with Cloudfleet. Cloudfleet selects these providers based on reliability, performance, service breadth, and customer demand. Cloudfleet currently supports the following CSPs for native node auto-provisioning:

1. Amazon Web Services (AWS): AWS offers Nitro-based EC2 instances for better security and performance. It also offers scalable Elastic Block Store (EBS) volumes up to 64 TiB and ultra-low-latency networking for HPC workloads.
2. Google Cloud Platform (GCP): GCP provides custom Tensor Processing Units (TPUs) for AI acceleration and high-performance Persistent Disks with up to 120,000 IOPS. Its global load balancing can handle millions of requests per second.
3. Hetzner Cloud: Hetzner delivers high-performance virtual machines with AMD EPYC and Intel Xeon processors, fast NVMe storage, and scalable private networking. Its cost-effective pricing suits developers and enterprises that need reliable European cloud infrastructure.

Learn how to configure Fleets here.

Cloudfleet continuously adds native integrations for more cloud providers, which expands service options and geographic coverage. If you want native support for a specific provider, let us know.

For infrastructure beyond these natively supported providers, you can use self-managed nodes to connect any cloud provider, on-premises infrastructure, or edge locations to CFKE. You can provision self-managed nodes automatically with Terraform or add them manually, so any infrastructure is supported immediately. For details, see the Self-managed nodes documentation.

Tuning Fleet behavior

Beyond connecting a cloud account, Fleets have several settings that control what infrastructure they provision and how they manage it:

• Fleet constraints: pin a Fleet to a specific cloud, region, instance family, architecture, or purchase type, so node auto-provisioning only considers matching infrastructure. Your Pod specifications stay simple.
• Fleet taints: taint every node of a Fleet, so only workloads that tolerate the taint run there. Use them to dedicate a Fleet to one kind of workload.
• Auto-provisioning profiles: control how aggressively the node auto-provisioner consolidates a Fleet, trading off cost optimization against workload disruption.

Supported accelerators

Cloudfleet offers a wide range of instance types with AI/ML accelerators. These accelerators are mostly specialized hardware such as GPUs, designed to increase the performance of machine learning workloads. Cloudfleet offers over 250 different instance types for the different requirements of AI and ML projects.

The Cloudfleet API uses standardized names for accelerator manufacturers and models. These standard names make it easier to identify and choose the right instance type for your needs.

The following table lists the supported accelerators and their available memory configurations:

Cloudfleet updates this list regularly as new accelerators become available on the platform. Check this page for the latest information and detailed specifications of each accelerator.