container.training/slides/k8s/kind.md

Working with KinD

• Kubernetes-in-Docker

• One of the many options to run local clusters

  (like Colima, Docker Desktop, k3d, Minikube, Podman Desktop...)

• Runs Kubernetes nodes inside Docker containers

  (also works with other container engines)

• Based on kubeadm

---

Pros

• Create a cluster in less than 1 minute

• Run multiple clusters side by side

• Clusters can have 1 or multiple nodes

• Select Kubernetes version to use

  (great to test regressions, upgrades...)

• Supports port mapping (more on that later)

• Very close to "upstream Kubernetes"

  (typically runs "normal" Kubernetes binaries)

---

Cons

• Requires a container engine

  (Docker, Podman...)

• Port mapping can only be configured at cluster creation time

  (more on that later)

• Releases can lag a bit behind Kubernetes "latest" releases

• Not a lot of pre-installed bells and whistles

  (e.g. no ingress controller, no metrics server...)

---

Getting started

• Creating a cluster:

  kind create cluster
  

• Cluster will be automatically added to your kubeconfig file

• See it running with docker ps

• Deleting a cluster:

  kind delete cluster
  

---

Customizing the cluster

• This is done with a configuration file

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
  image: kindest/node:v1.34.3
  extraPortMappings:
  - containerPort: 31234
    hostPort: 10000
- role: worker
  image: kindest/node:v1.34.3

• And then: kind create cluster --config XXX.yaml

---

Accessing internal apps

• There are many ways to access apps running in a KinD cluster

• We'll review some of them, with their pros/cons!

---

kubectl port-forward

• Works everywhere (local or remote, dev or prod...)

• Gated by RBAC with pod/port-forward subresource

• Inconvenient when accessing multiple services

• Inconvenient when working in Ingress or Gateway API

---

NodePorts

• Use NodePort services

• Plan A: access them with the node IP address

  does not work when the Docker host is a separate VM (e.g. with Docker Desktop)

• Plan B: configure KinD to map extra ports

  works everywhere; but needs some prep ahead of time

---

Example, mapping ports 80 and 443

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
  extraPortMappings:
  - containerPort: 30080
    hostPort: 80
  - containerPort: 30443
    hostPort: 443

• Configure Ingress or Gateway Controllers to use NodePorts 30080 and 30443

• Use a domain like localtest.me that resolves to 127.0.0.1

• Create Ingress or HTTPRoute resources using <something>.localtest.me

---

Variation

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
  extraPortMappings:
  - containerPort: 80
    hostPort: 80
  - containerPort: 443
    hostPort: 443

• Configure Ingress or Gateway Controllers to use hostNetwork or hostPort

---

Example, mapping a range of ports

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
  kubeadmConfigPatches:
  - |
    kind: ClusterConfiguration
    apiServer:
      extraArgs:
        service-node-port-range: "12301-12309"
  extraPortMappings:
  - { containerPort: 12301, hostPort: 12301 }
  - { containerPort: 12302, hostPort: 12302 }
  ...
  - { containerPort: 12308, hostPort: 12308 }
  - { containerPort: 12309, hostPort: 12309 }

(@@LINK[k8s/kind-12300.yaml])

---

Other noteworthy features

• Extra mounts to expose local files to pods

• Useful for big data sets (e.g. ML model weights)

• Need to remember that there is a double mapping:

  host → Kubernetes node in Docker container

  Kubernetes node in Docker container → pod

• Don't hesitate to use docker exec to get a shell in the Kubernetes node!

???

:EN:- Local dev clusters with KinD :FR:- Cluster de dev local avec KinD