Quickly replace a Google Cloud Kubernetes Cluster by a new one ( with new machine type, scopes and/or number of nodes ), keeping the credentials

replace-cluster.sh

#!/bin/bash

CLUSTER=""
PROJECT_CMD=""
ZONE_CMD=""
SCOPES=""
NUM_NODES="1"
MACHINE_TYPE=""
DISK_SIZE="100GB"

# Functions
function usage
{
    echo "usage: replace-cluster [ [-h] | --cluster CLUSTER --project PROJECT --machine-type TYPE [--num-nodes X(DEFAULT:1)] [--disk-size 100GB] [--scope SCOPE]... ] "
}

#run !
while [ "$1" != "" ]; do
    case $1 in
        -c | --cluster )        shift
                                CLUSTER=$1
                                ;;
        -p | --project )        shift
                                PROJECT_CMD="--project=${1}"
                                ;;
        -z | --zone )           shift
                                ZONE_CMD="--zone=${1}"
                                ;;
        --num-nodes )           shift
                                NUM_NODES="${1}"
                                ;;
        --machine-type )        shift
                                MACHINE_TYPE="${1}"
                                ;;
        --disk-size )           shift
                                DISK_SIZE="${1}"
                                ;;
        --scope )               shift
                                SCOPES="${SCOPES} ${1}"
                                ;;
        * )                     usage
                                exit
                                ;;
    esac
    shift
done

if [ "${CLUSTER}" == "" ]; then
    echo "Missing cluster name !"
    usage
    exit 1
fi
if [ "${MACHINE_TYPE}" == "" ]; then
    echo "Missing machine type !"
    usage
    exit 1
fi

if gcloud compute machine-types list ${PROJECT_CMD} ${ZONE_CMD} | grep -w ${MACHINE_TYPE} > /dev/null ; then
    echo "Using machine type ${MACHINE_TYPE}."
else
    echo "Machine type ${MACHINE_TYPE} not found."
    exit 1
fi

gcloud alpha container clusters get-credentials ${CLUSTER} ${PROJECT_CMD}

JSON_CLUSTER=$(gcloud alpha container clusters describe ${CLUSTER} ${PROJECT_CMD} --format=json)

SCOPES="$( (test -n "${SCOPES}" && echo ${SCOPES} | tr " " ","; echo ${JSON_CLUSTER} | jq -r ".nodeConfig.oauthScopes[] ") | paste -d ',' -s - )"


PASSWORD=$(echo ${JSON_CLUSTER} | jq -r .masterAuth.password)
USERNAME=$(echo ${JSON_CLUSTER} | jq -r .masterAuth.username)

echo "Scopes: " $SCOPES
echo "USERNAME: " $USERNAME
echo "PASSWORD: " $PASSWORD

TMPDIR=`mktemp -d /tmp/kubecfgs.XXXXXX` || exit 1

echo puting conf files in ${TMPDIR}


kubectl get rc --output=yaml > ${TMPDIR}/rc.yml
kubectl get services --output=yaml > ${TMPDIR}/services.yml

gcloud alpha container clusters delete ${PROJECT_CMD} ${ZONE_CMD} ${CLUSTER}

gcloud alpha container clusters create ${CLUSTER} ${PROJECT_CMD} --scopes ${SCOPES} --num-nodes ${NUM_NODES} --machine-type ${MACHINE_TYPE} --username=${USERNAME} --password=${PASSWORD}

gcloud alpha container clusters get-credentials ${CLUSTER} ${PROJECT_CMD}

kubectl create -f ${TMPDIR}/rc.yml
kubectl create -f ${TMPDIR}/services.yml

exit

With a bit of luck this keeps master ips.
With a huge bit of luck you can also keep a load balancer ingress ip.
( use at your own risk )