Justme.dev

Appearance

Converged Reliability: Strategic Hybrid Cloud with Harvester and Rancher

WatchersStarsForksGoShellOtherView on Github
Cloud-Native doesn't always mean "In the Cloud." During recent infrastructure projects, Harvester was used to transform bare-metal hardware into a private cloud, bridging the gap between legacy VMs and modern Kubernetes workloads with a single, unified control plane.

The Challenge: The "Legacy Island" Problem ​

Throughout a career in infrastructure, environments have been managed where modern Kubernetes clusters lived alongside "Legacy Islands". This created a dual-ops burden: one team for VMware/Nutanix and another for Kubernetes.

For mission-critical infrastructure, this operational fragmentation was a risk. A method was needed to manage VMs with the same GitOps and SRE principles used for containers.

The Strategy: Unified Hyperconverged Infrastructure (HCI) ​

Early experience in Integrating Physical Systems (AV, DMX, Camera systems) taught that every endpoint—no matter its age—should be addressable via a central API. Harvester was selected as the HCI foundation because it is built on Kubernetes, allowing for:

  1. Unified Control Plane: Manage both VMs and Containers side-by-side using Rancher.
  2. Kubernetes-Native Storage: Leveraging Longhorn for distributed, block-level storage across bare-metal nodes.
  3. Hardware Efficiency: Reclaiming underutilized resources by consolidating legacy VM workloads onto the same hardware as high-performance K8s clusters.

Implementation: Turning Bare Metal into a Cloud API ​

Harvester was deployed on commodity hardware, effectively building a private cloud region. This allowed the SRE team to use Terraform and Crossplane to provision VMs just as easily as provisioning S3 buckets in AWS.

yaml
# A strategic Harvester VM definition managed via Kubernetes APIs
apiVersion: devices.harvesterhci.io/v1beta1
kind: VirtualMachine
metadata:
  name: legacy-windows-service
spec:
  runStrategy: RerunOnFailure
  template:
    spec:
      domain:
        cpu: { cores: 4 }
        memory: { guest: 8Gi }
        devices:
          disks:
            - name: rootdisk
              bootOrder: 1
      volumes:
        - name: rootdisk
          containerDisk:
            image: harvester/os-images:windows-2022

By bringing the VM into the Kubernetes namespace, the same Cilium network policies and OpenTelemetry monitoring could be applied to legacy services as to microservices.

Impact: Operational Simplicity and 30% Better Utilization ​

The move to Harvester fundamentally changed the approach to hybrid-cloud:

  • Unified Operations: The SRE team now manages the entire lifecycle of both VMs and Containers through a single GitOps pipeline (ArgoCD + Rancher).
  • Hardware Rightsizing: Bare-metal hardware utilization was improved by 30% by eliminating the overhead of proprietary virtualization licenses and separate management stacks.
  • Edge Resilience: Harvester enables the operation of a "Cloud-in-a-Box" at remote edge locations, ensuring that critical crisis management tools remain available even if the connection to the central AWS region is severed.

Conclusion ​

Harvester represents the logical conclusion of the "Everything as Code" movement. Treating virtualization as just another Kubernetes workload removes the silos that slow down enterprise innovation.

Whether integrating DMX lighting or Enterprise SQL databases, the goal remains the same: to build a system that is unified, auditable, and resilient. Harvester is the foundation that makes this possible in a hybrid-cloud world.