Skip to main content
Version: 2.0.2

Overview

🏗️ Architecture Diagram​

High-Level Overview​

Hikube Kubernetes clusters rely on a multi-datacenter infrastructure (3 Swiss locations), ensuring replication, fault tolerance, and service continuity.

  • Control Plane: hosted and operated by Hikube Components:

    • kube-apiserver
    • etcd
    • kube-scheduler
    • kube-controller-manager

  • Worker Nodes: virtual machines inside your tenant

  • Networking: CNI with support for LoadBalancer, Ingress, and NetworkPolicy

  • Storage: persistent volumes replicated across the 3 datacenters

  • Add-ons: cert-manager, FluxCD, monitoring stack, etc.

  • Kubernetes Versioning: multi-version support with controlled upgrades

⚙️ Cluster Composition and Configuration​

Clusters are fully declarative and configurable via API or YAML manifests. The main configuration elements include:

ElementDescription
nodeGroupsHomogeneous groups of nodes (size, role, GPU, etc.)
storageClassDefines persistence and replication behavior
addonsOptional features that can be enabled
versionKubernetes server version
networkCNI configuration, LoadBalancer, Ingress

⚙️ How the Platform Works​

🧠 Control Plane​

  • Managed entirely by Hikube — no customer maintenance required
  • Critical components replicated across multiple sites
  • High availability, monitoring, and automated patching included
  • Access via the standard Kubernetes API (kubectl, SDK clients, etc.)

🧩 Worker Nodes / NodeGroups​

NodeGroups allow you to adapt compute resources to your needs. Each group can define instance type, roles, and autoscaling parameters.

Example NodeGroup​

nodeGroups:
  web:
    minReplicas: 2
    maxReplicas: 10
    instanceType: "s1.large"
    roles: ["ingress-nginx"]

Key Characteristics​

  • Autoscaling via minReplicas / maxReplicas
  • GPU support with dynamically attached NVIDIA GPUs
  • Instance types: S1 (standard), U1 (universal), M1 (memory-optimized)

💾 Persistent Storage​

Storage Class: replicated​

  • Automatic replication across all 3 Swiss datacenters
  • Dynamic provisioning of Persistent Volumes (PVC)
  • Built-in fault tolerance and high availability

Example usage:

storageClassName: replicated
resources:
  requests:
    storage: 20Gi

🔢 Kubernetes Versioning​

  • Clusters can be created with a specific Kubernetes version
  • Hikube handles minor and patch upgrades in a controlled manner
  • Customers may plan major upgrades when needed

Example:

version: "1.30.3"

🧩 Integrated Add-ons​

Cert-Manager​

  • Automated SSL/TLS certificate management
  • Supports Let’s Encrypt and private authorities
  • Automatic renewal

Ingress NGINX​

  • Built-in ingress controller
  • Wildcard support, SNI, and Prometheus metrics

Flux CD (GitOps)​

  • Continuous sync with your Git repositories
  • Automated deployments and rollback

Monitoring Stack​

  • Node Exporter, FluentBit, Kube-State-Metrics
  • Full integration with your tenant’s Grafana and Prometheus

🚀 Example Use Cases​

Web Applications​

nodeGroups:
  web:
    minReplicas: 2
    maxReplicas: 10
    instanceType: "s1.large"
    roles: ["ingress-nginx"]

ML/AI Workloads​

nodeGroups:
  ml:
    minReplicas: 1
    maxReplicas: 5
    instanceType: "u1.xlarge"
    gpus:
      - name: "nvidia.com/AD102GL_L40S"

Critical Applications​

nodeGroups:
  production:
    minReplicas: 3
    maxReplicas: 20
    instanceType: "m1.large"

📚 Resources​

💡 Key Takeaways​

  • Managed control plane – no master maintenance required
  • Workers in your tenant – full control over compute resources
  • Autoscaling – dynamic adjustment based on load
  • Multi-datacenter replication – built-in high availability
  • Full compatibility – standard Kubernetes API support