How to deploy a k8s cluster on VmWare ESX

k8s, vmware, ansible, metalLB, prometheus, grafana, alertmanager, metrics-server, harbor, clair, notary

If you’re short on resources in your home lab and simply can’t deploy an OpenStack private cloud to play around with, this tutorial will walk you through setting up a highly-available k8s cluster on VMware ESX vms. The principles used to deploy them very much resemble those used to deploy OpenStack instances so you’ll be able to quickly and effectively walk through the very first step in this tutorial.

The k8s cluster we’ll be deploying will have 3x master and 3x worker nodes and will use metalLB for bare-metal load balancing, prometheus, grafana and alertmanager for monitoring, metrics and reporting and harbor with clair and notary for image repository, signing and scanning.

We will be:

  • checking out and customising ansible-deploy-vmware-vm
  • creating vms from templates
  • setting up:
    • the k8s cluster via kubespray
    • metalLB
    • prometheus, grafana and alertmanager
    • metrics-server
    • harbor
    • notary and clair

What we won’t be doing is:

  • setting up VMwareESX (see here)
  • creating a VMware template (see here)

INFO: the VMware templates were made after an Ubuntu 18.04 vm.

STEP 1 — deploy multiple VMWare virtual machines from a template

Deploy the vm’s via Ansible. First and foremost, checkout ansible-deploy-vmware-vm and cd to it.

git clone
cd ansible-deploy-vmware-vm

Second, we need to tell ansible how to connect to our VMware ESX cluster. Edit or create the answerfile.yml and fill in the self explanatory blanks:

# - Defines the vCenter / vSphere environment
deploy_vsphere_host: '<vsphere_ip>'
deploy_vsphere_user: '<username>'
deploy_vsphere_password: '<password>'
deploy_vsphere_datacenter: '<datacenter>'
deploy_vsphere_folder: ''
esxi_hostname: '<esx_hostname>'

# Guest
# - Describes virtual machine common options
guest_network: '<network_name>'
guest_netmask: '<netmask>'
guest_gateway: '<gw>'
guest_dns_server: '<dns>'
guest_domain_name: '<domain_name>'
guest_id: '<guestID>'
guest_memory: '<RAM>'
guest_vcpu: '<CPU_cores>'
guest_template: '<template_name>'

Define the vms we want to deploy. Sample vms-to-deploy:

prod-k8s-master01 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Master #1'
prod-k8s-master02 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Master #2'
prod-k8s-master03 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Master #3'

prod-k8s-worker01 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Worker #01'
prod-k8s-worker02 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Worker #02'
prod-k8s-worker03 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Worker #03'

prod-k8s-harbor01 deploy_vsphere_datastore='ESX2' guest_custom_ip='<ip>' guest_notes='Harbor #01'

Define the playbook. Sample deploy-k8s-vms-prod.yml:

- hosts: all
  gather_facts: false
    - answerfile.yml
     - deploy-vsphere-template

Let’s deploy the vms:

ansible-playbook -vv -i vms-to-deploy deploy-k8s-vms-prod.yml


Deploy the k8s cluster on top of the created vms.

sudo pip install -r requirements.txt

# copy the sample inventory
cp -rfp inventory/sample inventory/mycluster

# declare all the k8s cluster IPs (masters and workers only)
declare -a IPS=(x.x.x.x y.y.y.y z.z.z.z)
# see below samples before running this cmd
CONFIG_FILE=inventory/mycluster/hosts-prod.yml python3 contrib/inventory_builder/ ${IPS[@]}

sample hosts-prod.yaml:

      ansible_host: <ip>
      ip: <ip>
      access_ip: <ip>
      ansible_host: <ip>
      ip: <ip>
      access_ip: <ip>
      ansible_host: <ip>
      ip: <ip>
      access_ip: <ip>
      ansible_host: <ip>
      ip: <ip>
      access_ip: <ip>
      ansible_host: <ip>
      ip: <ip>
      access_ip: <ip>
      ansible_host: <ip>
      ip: <ip>
      access_ip: <ip>
      hosts: {}

sample inventory.ini (make sure the inventory file contains the vm’s proper name (ie. the ones defined under “ansible-deploy-vmware-vm/vms-to-deploy”))

# ## Configure 'ip' variable to bind kubernetes services on a
# ## different ip than the default iface
# ## We should set etcd_member_name for etcd cluster. The node that is not a etcd member do not need to set the value, or can set the empty string value.
# node1 ansible_host=  # ip= etcd_member_name=etcd1
# node2 ansible_host=  # ip= etcd_member_name=etcd2
# node3 ansible_host=  # ip= etcd_member_name=etcd3
# node4 ansible_host=  # ip= etcd_member_name=etcd4
# node5 ansible_host=  # ip= etcd_member_name=etcd5
# node6 ansible_host=  # ip= etcd_member_name=etcd6

# ## configure a bastion host if your nodes are not directly reachable
# bastion ansible_host=x.x.x.x ansible_user=some_user

# node1
# node2

# node1
# node2
# node3

# node2
# node3
# node4
# node5
# node6


Run the playbook to create the cluster:

ansible-playbook -vvv -i inventory/mycluster/hosts-prod.yml --become --become-user=root cluster.yml


MetalLB is a bare-metal load balancer for k8s that makes your current network extend into your k8s cluster.

Connect to one of the master nodes over ssh and setup your environment:

source <(kubectl completion bash) # or zsh

Deploy metalLB by performing the following on one of the master nodes:

kubectl apply -f

MetalLB remains idle until configured. As such, we need to define an IP range the k8s cluster can use and that is outside of any DHCP pool.

cat <<EOF | kubectl apply -f - 
apiVersion: v1
kind: ConfigMap
  namespace: metallb-system
  name: config
  config: |
    - name: default
      protocol: layer2
      - <ip_range_start>-<ip_range_end>

Check the status of the pods with:

kubectl -n metallb-system get po


Install prometheus, grafana and alertmanager:

git clone
cd kube-prometheus
kubectl create -f manifests/setup
until kubectl get servicemonitors --all-namespaces ; do date; sleep 1; echo ""; done
kubectl create -f manifests/

# To teardown the stack:
#kubectl delete -f manifests/

In order to access the dashboards via the LoadBalancer IPs, we need to change a few service types from “ClusterIP” to “LoadBalancer”:

kubectl -n monitoring edit svc prometheus-k8s
kubectl -n monitoring edit svc grafana
kubectl -n monitoring edit svc alertmanager-main

# Get the external IP for the edited services
kubectl -n monitoring get svc

In order to access the web interface of these services, use the following ports:

– grafana -> 3000 (default usr/pass is admin/admin)
– prometheus -> 9090
– alertmanager -> 9093


Installing the metrics-server or how to get “kubectl top nodes” and “kubectl top pods” to work.

git clone
cd metrics-server
kubectl create -f deploy/1.8+/

kubectl top nodes
kubectl top pods --all-namespaces


Installing Harbor with clair and notary support.

ssh to your harbor vm and install docker and docker-compose:

curl -fsSL | sudo apt-key add -

sudo add-apt-repository \
   "deb [arch=amd64] \
   $(lsb_release -cs) \

sudo apt-get update && sudo apt-get install -y docker-ce
sudo curl -L "$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
sudo chmod +x /usr/local/bin/docker-compose

# allow  users to use docker without administrator privileges
sudo usermod -aG docker $USER

Logout and login and check docker:

docker info

Time to generate SSL certificated:

# generate a certificate authority
openssl req -newkey rsa:4096 -nodes -sha256 -keyout ca.key -x509 -days 3650 -out ca.crt

# generate a certificate signing request
openssl req -newkey rsa:4096 -nodes -sha256 -keyout prod-k8s-harbor01.lab.local.key -out prod-k8s-harbor01.lab.local.csr

# Create a configuration file for the Subject Alternative Name.
vim extfile.cnf
subjectAltName = IP:<ip>

# generate the certificate
openssl x509 -req -days 3650 -in prod-k8s-harbor01.lab.local.csr -CA ca.crt -CAkey ca.key -CAcreateserial -extfile extfile.cnf -out prod-k8s-harbor01.lab.local.crt

# copy the certificate to /etc/ssl/certs.
sudo cp *.crt *.key /etc/ssl/certs

Download and install Harbor:

tar xvzf harbor-online-installer-v1.9.4.tgz
cd harbor

Edit harbor.yml and change a few options:

hostname: <harbor_ip>

# http related config
  port: 80
  port: 443
  certificate: /etc/ssl/certs/prod-k8s-harbor01.lab.local.crt
  private_key: /etc/ssl/certs/prod-k8s-harbor01.lab.local.key
harbor_admin_password: <your_admin_pass>

Finally, install Harbor, enabling clair and notary:

sudo ./ --with-notary --with-clair

Configure the docker daemon on each of your worker nodes and

export harbor_ip=<harbor_local_ip>
declare -a IPS=(<k8s_master1> <k8s_master2> <k8s_master3>)
for i in ${IPS[@]}; do
  scp ../ca.crt $i:
  ssh $i "sudo mkdir -p /etc/docker/certs.d/$harbor_ip && \
    sudo mv ca.crt /etc/docker/certs.d/$harbor_ip/ && \
    sudo systemctl restart docker"

Login to a k8s master node and create a secret object for harbor:

kubectl create secret docker-registry harbor \
--docker-server=https://<harbor_ip> \
--docker-username=admin \ \

Login to the harbor web-interface and create a new repository called “private”.

To deploy images to Harbor, we need to pull them, tag them and push them. From the harbor machine (or any other that has the certificates in place as above), perform the following:

docker pull
docker tag <harbor_ip>/private/kuard:v1
docker login <harbor_ip>
docker push <harbor_ip>/private/kuard:v1

Deploy the kuard app on k8s. ssh to a node where you have access to the cluster and create kuard-deployment.yaml:

apiVersion: apps/v1
kind: Deployment
  name: kuard-deployment
    app: kuard
  replicas: 1
      run: kuard
        run: kuard
      - name: kuard
        image: <harbor_ip>/private/kuard:v1
      - name: harbor

Apply it.

kubectl create -f kuard-deployment.yaml
kubectl get po


Signing docker images with Notary. Install it first:

sudo wget -O /usr/local/bin/notary
sudo chmod +x /usr/local/bin/notary

If needed, copy the “ca.crt” to the client machine you’re working from.

Check if you can connect to the harbor server:

openssl s_client -connect <harbor_ip>:443 -CAfile /etc/docker/certs.d/<harbor_ip>/ca.crt -no_ssl2

You should get something like this:

CONNECTED(00000005)depth=1 C = EU, ST = Example, L = Example, O = Example, OU = Example, CN = ca.example.local, emailAddress = root@ca.example.localverify return:1depth=0 C = EU, ST = Example, O = Example, CN = notary-server.example.localverify return:1---Certificate chain0 s:/C=EU/ST=Example/O=Example/CN=notary-server.example.locali:/C=EU/ST=Example/L=Example/O=Example/OU=Example/CN=ca.example.local/emailAddress=root@ca.example.local

Now pull an image from docker hub and tag it but don’t push it just yet.

docker pull nginx:latest
docker tag nginx:latest <harbor_ip>/private/nginx:latest

Let’s enable the Docker Content Trust and then push the image. Please note that when first pushing a signed image, you will be asked to create a password.

docker login <harbor_ip>
docker push <harbor_ip>/private/nginx:latest

Ok, so now the Docker image is pushed in our Registry server and it is signed by the Notary server. We can verify

notary --tlscacert ca.crt -s https://<harbor_ip>:4443 -d ~/.docker/trust list <harbor_ip>/private/nginx

Test that you can pull from docker hub and harbor:

docker pull nginx
docker pull <harbor_ip>/private/nginx:1.16.0

For using a completely private repository, leave the “DOCKER_CONTENT_TRUST_SERVER” and the “DOCKER_CONTENT_TRUST” environment variables set, on all the k8s cluster machines.


Clair is an open source project for the static analysis of vulnerabilities in application containers. To enable it go to the Harbor web-interface and click on “vulnerability” -> “edit” and select the scan frequency. Save and also click “scan now” if it’s your first time doing this.

ESX 6.5 and 6.7 on HPE G6/G7 server PSOD fix


HP’s done it again. They’ve managed to break their custom ESX ISO on the G6 and G7 servers. I suspect it’s the same for G8.

If like me you found yourself perplexed by the PSOD (pink screen of death) after upgrading from ESX 6.0 to 6.5, keep on reading for the fix (skip to the bottom for the download link to an already fixed ISO).

The issue

It seems the “hpe-smx-provider” driver version 6.5.0 from the ESX 6.5 ISO is causing the PSOD.

The fix and the drawbacks

(Simply) Replace the driver with it’s older counterpart (version 6.0.0) and reinstall (see below for the procedure or check the “download link” section at the very end for an already fixed ISO).  Unfortunately if you decide to create a custom ISO with the previous version of “hpe-smx-provider”, you will no longer be able to upgrade to any future ESX version and will need to do a full installation every time (thank you HP!). That being said, here is the procedure to customize the ISO:

Customize your own HPE ESX ISO

  1. Install PowerCLI from here.  At the time of this writing, the latest version was 6.5.0R1.
  2. Download both the ESX 6.5 and 6.0 offline bundles and save them to a convenient place (ex: C:\HP) — download from this link. To make it easier, rename them to something like and
  3. Open up PowerCLI and do the following:Add the 6.5 bundle
    cd C:\HP
    Add-EsxSoftwareDepot -DepotUrl

    # check that the profile was loaded and the vendor is HPE


    # clone the profile (feel free to specify whatever you like for the “vendor” when prompted)

    New-EsxImageProfile -CloneProfile HPE-ESXi-6.5.0* -Name "HPE-ESX-6.5-CUST"

    # double check the profile was indeed cloned


    # Remove the “hpe-smx-provider” driver from the clone

    Remove-EsxSoftwarePackage HPE-ESX-6.5-CUST hpe-smx-provider

    # Add the 6.0 bundle

    Add-EsxSoftwareDepot -DepotUrl

    # check the profile


    # check driver versions from both bundles

    Get-EsxSoftwarePackage | findstr smx

    # Add hpe-smx-provider version 6.00 to the custom profile

    add-esxsoftwarepackage -imageprofile HPE-ESX-6.5-CUST -softwarepackage "hpe-smx-provider 600."

    # Export the profile to an ISO

    Export-EsxImageProfile -ImageProfile HPE-ESX-6.5-CUST -ExportToIso -filepath "HPE-ESX-6.5-CUST.iso"

The results

Download links

The following images have hpe-smx-provider version 600. Nothing else was changed. The 6.5.0d (build 5310538) is running on an HP Proliant DL380 G7 since March 2018 and was also tested on a DL380 G6 for a couple of months. For convenience, I’ve also built a 6.7.0 ISO and at the request of a reader, 6.7.0 Update1 from April of 2019: