Skip to content
This repository has been archived by the owner on Feb 5, 2020. It is now read-only.

Latest commit

 

History

History
75 lines (56 loc) · 3.38 KB

generic-platform.md

File metadata and controls

75 lines (56 loc) · 3.38 KB
Raw

Generic Platform

This document describes the generic compute, network, and storage needs of a Kubernetes self-hosted cluster. In particular the master, infra, and worker nodes.

Key User Question: What is the minimum amount of compute/network architecture that is required to enable automated operations that a user cannot change?

DNS Setup

  • API server will generally be <cluster-name>-k8s.<subdomain>
  • Console, Dex, etc will generally be <cluster-name>.<subdomain>

Nodes

Master Nodes

Master nodes run most, if not all, control plane components including the API server, etcd, scheduler, controller manager, etc.

  • Count: At least one dedicated master node for etcd and Kubernetes components, 3 or 5 nodes for HA

  • Network:

    • Ingress
      • MUST allow tcp port 22 (ssh) from user network
      • MUST allow port 4789 (UDP) from masters & workers for flannel
      • MUST allow 32000-32002 from all for: Tectonic ingress (if using node ports for ingress like on AWS, otherwise use host ports on workers)
      • SHOULD allow port 9100 from masters & workers for: Prometheus Node Exporter metrics
      • MAY have tcp/udp port 30000-32767 (node port range open)
      • MAY have tcp port 2379 (etcd client API when not using external etcd)
      • MAY allow port 10255 from all for: read-only kubelet status & LB health checks for k8s API
    • Egress
      • MUST have 443 to download gcr, quay, and docker hub images
      • MAY have 2379 to external etcd cluster
      • MAY allow 2379-2380 for self-hosted etcd pods if using experimental etcd operator
      • MAY allow 12379-12380 for temporary etcd pod if using experimental etcd operator

  • Access Control

    • MUST have any necessary API access for k8s cloud plugin functionality (i.e. AWS node IAM Role)

  • Cloud-init/Ignition:

    • MUST run the kubelet with label node-role.kubernetes.io/master
    • MAY run the kubelet with label with label node-role.kubernetes.io/node

Infra Nodes (Optional) or Load Balancers

Infra nodes are special nodes that must be deployed when a load balancer isn't available in an environment.

  • Count: 0 or more
  • Network:
    • Ingress
      • MUST have tcp port 443 (Ingress controller for Tectonic Console and Dex)
    • Egress
      • MUST have 443 to download gcr, quay, and docker hub images
  • Cloud-init/Ignition:
    • MUST run the kubelet
    • MAY have a taint annotation to avoid user workloads from running here

Worker Nodes

Worked nodes run all of the user applications. The only component they must run on-boot is the kubelet.

  • Count: User specified N

  • Network:

    • Ingress

      • MUST allow all ports open to master nodes (TODO: be more specific)
      • MUST have 30000 to 32767 host port range access open
      • MUST allow port 4789 (UDP) from masters & workers for: VXLAN (flannel)
      • SHOULD allow port 10250 from masters for k8s features: port-forward, exec, proxy
      • SHOULD allow port 9100 from masters & workers for: Prometheus Node Exporter metrics
      • SHOULD allow port 4194 from masters for: Heapster connections to CAdvisor
      • MAY allow 2379-2380 for self-hosted etcd pods if using experimental etcd operator

    • Egress

      • MUST have 443 to download gcr, quay, and docker hub images

  • Cloud-init/Ignition

    • MUST run the kubelet with label node-role.kubernetes.io/node=true