Skip to content

Supported Environments

Supported environments and operational requirements for upstream Kubernetes (MKS) on bare metal and VMs.


Operating Systems

Operating System Control plane Worker nodes
AlmaLinux 9 (64-bit) Yes Yes
RHEL 8.x (64-bit) Yes Yes
RHEL 9.1 and 9.2 (64-bit) Yes Yes
RHEL 10 (64-bit) Yes Yes
Rocky Linux 9 (64-bit) Yes Yes
Ubuntu 20.04 LTS (64-bit) (EOL) Yes Yes
Ubuntu 22.04 LTS (64-bit) Yes Yes
Ubuntu 24.04 LTS (64-bit) Yes Yes
Windows Server 2019 (64-bit) No Yes
Windows Server 2022 (64-bit) No Yes
Flatcar Linux 4081 Yes Yes

RHEL kernel

On RHEL-based systems, kernel 4.18 is not supported with Kubernetes 1.33 and above. Kernel 4.19 is the recommended 4.x LTS option; 5.x and 6.x kernels are preferred for Kubernetes 1.33+.

Windows workers

Windows worker nodes require Kubernetes v1.23.x or higher and the Calico CNI. Canal and Cilium are not supported with Windows workers.


Hypervisors

MKS on VMs is hypervisor-agnostic. The following have been validated for provisioning and lifecycle management:

  • VMware vSphere (v7.x, v8.x)
  • Microsoft Hyper-V
  • Nutanix AOS (v6.5.x LTS, v6.8.x)
  • OpenStack (2023.1 Antelope, 2023.2 Bobcat, 2024.1 Caracal)
  • VirtualBox (v7.0.x, v6.1.x)

Kubernetes versions

Rafay supports four minor Kubernetes versions at a time. When a new minor version is added, the oldest supported minor version is removed. Upgrade clusters regularly to stay on supported versions.

Supported minor versions

Version End of standard support Added in controller release
v1.36.x N/A v4.1 Update 7
v1.35.x 25 Feb 2027 v4.0 Update 4
v1.34.x 27 Oct 2026 v3.7
v1.33.x 28 Jun 2026 v3.5
v1.32.x (deprecated) 28 Feb 2026 v3.1
v1.31.x (EOL) 28 Oct 2025 v2.10
v1.30.x (EOL) 28 Jun 2025 v2.7
v1.29.x (EOL) 28 Feb 2025 v2.4
v1.28.x (EOL) 28 Oct 2024 v2.0
v1.27.x (EOL) 28 Jun 2024 v1.27
v1.26.x (EOL) 28 Feb 2024 v1.25
v1.25.x (EOL) 27 Oct 2023 v1.19
v1.24.x (EOL) 28 Jul 2023 v1.15
v1.23.x (EOL) 28 Feb 2023 v1.11

Note

Kubernetes 1.32.x reaches upstream end of life on 28 Feb 2026. Rafay will continue to support it for a transition period so customers can migrate to a supported minor version.

Patch versions

Default patches are shown in the UI when provisioning or upgrading. Deprecated patches are labeled (deprecated) and appear when Show deprecated Kubernetes patch versions is enabled.

Minor Default patch Deprecated patches
v1.36 v1.36.0
v1.35 v1.35.4 v1.35.0
v1.34 v1.34.7 v1.34.3, v1.34.1
v1.33 v1.33.11 v1.33.7, v1.33.5
v1.32 v1.32.13, v1.32.11, v1.32.9

Use the default patch for each minor version (for example, v1.35.4 for 1.35). Deprecated patches are for migration and upgrade testing only.


Platform versioning

Each platform version bundles cluster components (CRI, etcd, agents) for consistent upgrades. Platform version v1.3.0 is the latest and the default for new cluster creation.

Component versions

Component v1.3.0 (latest) v1.2.0 v1.1.0 (deprecated)
CRI 2.3.0 2.0.4 2.0.4
etcd 3.5.24 3.5.24 3.5.21
Cluster Utils 2.0.0 2.0.0 2.0.0
Orchestration Agent 3006.12 3006.12 3006.12
Orchestration Proxy 1.10.1 1.10.1 1.10.1

Platform version summary

Version Status Notes
v1.3.0 Latest (default) Default for new cluster creation; required for Kubernetes v1.36. CRI 2.3.0 is the minimum containerd version for v1.36.
v1.2.0 Active Required for Kubernetes v1.35 and default patches v1.35.4, v1.34.7, v1.33.11. etcd 3.5.24 is required for Kubernetes 1.35+.
v1.1.0 Deprecated Migration only. See component table above.
v1.0.0 Deprecated Migration only.
v0.1.0 Deprecated Reference label for clusters created before platform versioning. Upgrade to v1.2.0 (direct upgrade from v0.1.0 is supported).

Deprecated platform versions (v1.1.0, v1.0.0, v0.1.0) are labeled (deprecated) in the UI. Enable Show deprecated platform versions to select them.


Node management and cluster upgrades

Prerequisites

  • Control plane: 8 OCPUs (16 vCPUs), 32 GB memory
  • Workers: 1 OCPU (2 vCPUs), 4 GB memory

Recommendations

  • Add or remove nodes in batches of up to 100
  • If some nodes fail during upgrade, the retry mechanism upgrades the remainder

Important

Qualified for up to 500 nodes and 10,000 pods.


Container networking (CNI)

CNI Notes
Cilium Recommended for Linux nodes
Calico Recommended for Linux and Windows nodes
Canal Calico + Flannel
Flannel Deprecated; not recommended for new clusters

CPU, memory, and architecture

  • Control plane: Linux/x64 and Linux/arm64
  • Workers: Linux/x64, Linux/arm64, or Windows/x64
  • Full ARM support on Ubuntu 22.04 and 24.04 LTS (default and default-upstream blueprints)

Minimal blueprint

Resource Minimum
vCPUs per node 2
Memory per node 4 GB

default-upstream blueprint

Single-node cluster:

Resource Minimum Cores
vCPUs per node 2 4
Memory per node 16 GB

HA cluster:

Resource Minimum Cores
vCPUs per node 2 4
Memory per node 16 GB

Important

Allow extra capacity if you deploy blueprints with monitoring, storage, or other add-ons. To switch from default-upstream to another blueprint after provisioning, delete workload deployments and workload PVCs first.


GPU

NVIDIA GPUs compatible with Kubernetes are supported. See GPU overview.


Container runtime

Clusters use containerd (CRI) instead of Dockershim (removed from Kubernetes 1.20+). In-place Kubernetes upgrades also move nodes to containerd; plan for workload restarts.


Inter-node networking

Multi-node clusters require node-to-node connectivity across all UDP/TCP ports.

Control plane

Protocol Direction Ports Purpose
TCP Inbound 6443 API server
TCP Inbound 2379-2380 etcd
TCP Inbound 10250, 10255 kubelet
TCP Inbound 10259, 10251 scheduler
TCP Inbound 10257, 10252 controller-manager
UDP Inbound 8285 Flannel
TCP Inbound 30000-32767 NodePort (if exposed on control plane)
TCP Inbound 9099 Calico
TCP Inbound 5656 OpenEBS Local PV
UDP Inbound 4789 VXLAN

Workers

Protocol Direction Ports Purpose
TCP Inbound 10250, 10255 kubelet
TCP Inbound 30000-32767 NodePort
UDP Inbound 8285, 8472 Flannel
TCP Inbound 8500 Consul
UDP Inbound 8600 Consul
TCP/UDP Inbound 8301 Consul
TCP Inbound 9099 Calico
TCP Inbound 5656 OpenEBS Local PV
UDP Inbound 4789 VXLAN

Forward proxy

Configure when nodes cannot reach the controller directly and must use a non-transparent proxy.


Storage

Turnkey storage options in the cluster blueprint simplify PersistentVolume management on bare metal and VMs.

Local PV

Required storage class.

  • OpenEBS on bare metal and VMs
  • Amazon EBS on EC2 (requires IAM role for dynamic provisioning)

Suited for: workloads with built-in replication (MongoDB, Redis, Cassandra, Postgres), high local throughput, or single-node/edge clusters without shared storage.

Distributed storage

Optional Rook-Ceph for highly available shared storage and pod rescheduling across workers.

Important

GlusterFS-based managed storage was deprecated in Q1 2022 and EOL in Q1 2023.

Storage requirements

Root disk

Used for container images, Kubernetes binaries, etcd, Consul, system packages, and logs (rotated via logrotate).

  • Raw, unformatted
  • Minimum 50 GB; recommended 100 GB+

Note

On a single-node cluster, reserve about 30 GB for platform data and 20 GB for workload PVCs unless you plan additional capacity.

Secondary disk

Optional; required only for Rook-Ceph. Dedicated to workload PVCs.

  • Raw, unformatted
  • Minimum 50 GB per node; recommended 100 GB+