# OCI Service: Oracle Container Engine for Kubernetes (OKE)
# Last verified: 2026-03

service:
  name: "Oracle Container Engine for Kubernetes (OKE)"
  id: oke
  category: infrastructure
  color: "#2D5967"

description: "Managed Kubernetes service with Oracle-optimized control plane, integrated with OCI networking, IAM, and observability."

when_to_use:
  - "Containerized applications (microservices, 12-factor apps)"
  - "Need horizontal auto-scaling for application tier"
  - "Multi-team environments needing namespace isolation"
  - "CI/CD-driven deployments"
  - "Workloads requiring service mesh, ingress controllers"

when_not_to_use:
  - "Simple stateless functions → use OCI Functions instead"
  - "Single monolithic app with no container strategy → use Compute VMs"
  - "Team has no Kubernetes experience and timeline is tight"
  - "Batch processing only → consider OCI Data Flow or Compute"

variants:
  - id: oke_managed
    name: "OKE Managed Nodes"
    description: "Customer manages worker nodes (OS patching, scaling config)"
  - id: oke_virtual
    name: "OKE Virtual Nodes"
    description: "Serverless worker nodes — no OS management, pay per pod"
    use_when:
      - "Want zero node management overhead"
      - "Bursty workloads with variable pod count"
      - "Cost optimization for intermittent workloads"

sizing:
  control_plane: "Free (managed by Oracle)"
  worker_nodes:
    recommended_shapes:
      - "VM.Standard.E4.Flex (general purpose, best price/performance)"
      - "VM.Standard.A1.Flex (ARM/Ampere, 30-50% cheaper for compatible workloads)"
      - "VM.Standard3.Flex (Intel, when Intel compatibility required)"
      - "VM.GPU.A10.1 / VM.GPU.A100 (GPU workloads)"
    sizing_guidance: "Start with 3 nodes across fault domains for HA. Size OCPUs based on aggregate pod CPU requests + 20% headroom."

gotchas:
  - id: oke_lb_shape
    severity: LOW
    description: "OKE creates OCI Load Balancers for LoadBalancer-type services. Default is 10 Mbps flexible — adjust annotation for production bandwidth."

  - id: oke_cni_choice
    severity: MEDIUM
    description: "VCN-Native Pod Networking (OCI CNI) gives pods VCN IPs directly — better for NSG integration but consumes subnet IPs. Flannel overlay is simpler but no direct VCN integration."

  - id: oke_pv_block_volume
    severity: LOW
    description: "Persistent volumes use OCI Block Volumes. For ReadWriteMany, use OCI File Storage (FSS) CSI driver."

implied_dependencies:
  - service: vcn
    reason: "Worker nodes and pods need VCN subnets"
  - service: load_balancer
    reason: "Ingress traffic to services"
  - service: block_storage
    reason: "Persistent volumes for stateful pods"

references:
  documentation: "https://docs.oracle.com/en-us/iaas/Content/ContEng/home.htm"

changelog:
  - date: "2026-03-14"
    contributor: { name: "Diego Cabrera", team: "Field Architecture" }
    change: "Initial creation with managed/virtual nodes, sizing, gotchas, dependencies"