upgrade kind kubernetes cluster

README.md

KIND development workflow

1. Create your kind cluster (suit yourself https://kind.sigs.k8s.io/docs/user/configuration/)

cat <<EOF | kind create cluster --config=-
kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
- role: worker
- role: worker
EOF

2. Work in your Kubernetes branch, and when you are ready, deploy your changes in the running cluster

[aojea@myhost kubernetes]$ ./kind_upgrade.sh 
Updated component kube-apiserver on node kind-control-plane
Updated component kube-controller-manager on node kind-control-plane
Updated component kube-scheduler on node kind-control-plane
Updated kubelet on node kind-control-plane
Updated kubelet on node kind-worker2
Updated kubelet on node kind-worker
daemon set "kube-proxy" successfully rolled out
Updated kube-proxy
KIND cluster kind updated successfully with version v1.20.0-alpha.2.199+88d193939d072c-dirty

The script by default, will upgrade all the components in all the nodes with your local changes,

You can tweak this behavior, and update only some components with the different flags.

E2E tests

If you want to run the e2e tests, you just have to compile the e2e binary and run against the cluster. You can find more detailed instructions in:

https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/conformance-tests.md#running-conformance-tests-with-kind

kind_upgrade.sh

#!/usr/bin/env bash
set -e
set -o pipefail

build_docker(){
  build/run.sh make all WHAT="cmd/kubectl cmd/kubelet" 1> /dev/null
  make quick-release-images 1> /dev/null
}

build_bazel(){
  bazel build //cmd/kubectl:kubectl //cmd/kubelet:kubelet //build:docker-artifacts
}

update_kubelet() {
 for n in $NODES; do
   # Backup previous kubelet
   docker exec $n cp /usr/bin/kubelet /usr/bin/kubelet.bak
   # Install new kubelet binary
   docker cp ${KUBELET_BINARY} $n:/usr/bin/kubelet
   docker exec $n systemctl restart kubelet
  echo "Updated kubelet on node $n"
 done
}

update_kube_proxy() {
  for n in $NODES; do
   kind load image-archive ${IMAGES_PATH}/kube-proxy.tar --name ${CLUSTER_NAME}
  done
  # RollingUpdate
  kubectl set image ds/kube-proxy kube-proxy=${DOCKER_REGISTRY}/kube-proxy-amd64:${DOCKER_TAG} -n kube-system
  kubectl rollout status ds/kube-proxy -n kube-system -w
  echo "Updated kube-proxy"
}

update_cni() {
  for n in $NODES; do
   kind load image-archive ${CNI_IMAGE} --name ${CLUSTER_NAME}
  done
  # RollingUpdate
  kubectl set image ds/kindnet kindnet-cni=${CNI_IMAGE} -n kube-system
  kubectl rollout status ds/kindnet -n kube-system -w
  echo "Updated kindnet"
}

update_control_plane(){
  # TODO allow to configure node and control plane components
  for n in $CONTROL_PLANE_NODES; do
    for i in $CONTROL_PLANE_COMPONENTS; do
      kind load image-archive ${IMAGES_PATH}/${i}.tar --name ${CLUSTER_NAME} --nodes $n
      docker exec $n sed -i.bak -r "s|^(.*image\:.*)\:.*$|\1-amd64\:${DOCKER_TAG}|" /etc/kubernetes/manifests/$i.yaml
      echo "Updated component $i on node $n"
      sleep 1
    done
  done
}

usage()
{
    echo "usage: kind_upgrade.sh [-n|--name <cluster_name>] [--cni <cni_image>] [-b|--build-mode docker|bazel]"
    echo "                       [--no-kproxy]  [--no-control-plane]  [--no-kubelet]"
    echo ""
}

parse_args()
{
    while [ "$1" != "" ]; do
        case $1 in
            -n | --name )			shift
                                          	CLUSTER_NAME=$1
                                          	;;
            --cni-image )	           	shift
                                          	CNI_IMAGE=$1
                                          	;;
            -b | --build-mode )			shift
                                                if [ "$1" != "docker" ] && [ "$1" != "bazel" ]; then
                                                    echo "Invalid build mode: $1"
                                                    usage
                                                    exit 1
                                                fi
                                                BUILD_MODE=$1
                                                ;;
            --no-kproxy )                   	UPDATE_KUBE_PROXY=false
                                                ;;
            --no-kubelet )                   	UPDATE_KUBELET=false
                                                ;;
            --no-control-plane )               	UPDATE_CONTROL_PLANE=false
                                                ;;
            -h | --help )                       usage
                                                exit
                                                ;;
            * )                                 usage
                                                exit 1
        esac
        shift
    done
}

parse_args $*

# Set default values
CLUSTER_NAME=${CLUSTER_NAME:-kind}
BUILD_MODE=${BUILD_MODE:-docker}
UPDATE_KUBE_PROXY=${UPDATE_KUBE_PROXY:-true}
UPDATE_KUBELET=${UPDATE_KUBE_PROXY:-true}
# TODO: we can have more granularity here
UPDATE_CONTROL_PLANE=${UPDATE_CONTROL_PLANE:-true}
CONTROL_PLANE_COMPONENTS="kube-apiserver kube-controller-manager kube-scheduler"

# Initialize variables
# Assume go installed
KUBE_ROOT="$(go env GOPATH)/src/k8s.io/kubernetes"
source "${KUBE_ROOT}/hack/lib/version.sh"
kube::version::get_version_vars
DOCKER_TAG=${KUBE_GIT_VERSION/+/_}
DOCKER_REGISTRY=${KUBE_DOCKER_REGISTRY:-k8s.gcr.io}
export GOFLAGS="-tags=providerless"
export KUBE_BUILD_CONFORMANCE=n

# KIND nodes
NODES=$(kind get nodes --name ${CLUSTER_NAME})
CONTROL_PLANE_NODES=$(kind get nodes --name ${CLUSTER_NAME} | grep control)
WORKER_NODES=$(kind get nodes --name ${CLUSTER_NAME} | grep worker)

# Main
if [[ "$BUILD_MODE" == "docker" ]]; then
  build_docker
  IMAGES_PATH="${KUBE_ROOT}/_output/release-images/amd64"
  KUBELET_BINARY=$(find ${KUBE_ROOT}/_output/ -type f -name kubelet)
else
  build_bazel
  IMAGES_PATH="${KUBE_ROOT}/bazel-kubernetes/bazel-out/k8-fastbuild/bin/build"
  KUBELET_BINARY=$(find ${KUBE_ROOT}/bazel-kubernetes/ -type f -name kubelet)
fi

if [[ "$UPDATE_CONTROL_PLANE" == "true" ]]; then
  update_control_plane
fi

if [[ "$UPDATE_KUBELET" == "true" ]]; then
  update_kubelet
fi

if [[ "$UPDATE_KUBE_PROXY" == "true" ]]; then
  update_kube_proxy
fi

# If CNI_IMAGE set update the CNI  
if [[ ! -z ${CNI_IMAGE} ]]; then
  update_cni
fi

if kubectl get nodes | grep NoReady; then
  echo "Error: KIND cluster $CLUSTER_NAME NOT READY"
  exit 1
else
  echo "KIND cluster $CLUSTER_NAME updated successfully with version $KUBE_GIT_VERSION"
  exit 0
fi

Doesn't work if cluster already exists:

$bash -x upgrade-kind.sh
+ set -o errexit
+ reg_name=kind-registry
+ reg_port=5001
++ docker inspect -f '{{.State.Running}}' kind-registry
+ '[' true '!=' true ']'
+ cat
+ kind create cluster --config=-
ERROR: failed to create cluster: node(s) already exist for a cluster with the name "kind"

kind create cluster or kind-registry is not part of the script, how are those failing in the script?

L106 should be:

UPDATE_CONTROL_PLANE=${UPDATE_CONTROL_PLANE:-true}

Thanks @sanposhiho for catching it :)