I've been leveraging the open-source Tanzu Community Edition Kubernetes distribution for a little while now, both [in my home lab](/tanzu-community-edition-k8s-homelab) and at work, so I was gutted by the news that VMware was [abandoning the project](https://github.com/vmware-tanzu/community-edition). TCE had been a pretty good fit for my needs, and now I needed to search for a replacement. VMware is offering a free version of Tanzu Kubernetes Grid as a replacement, but it comes with a license solely for non-commercial use so I wouldn't be able to use it at work. And I'd really like to use the same products in both environments to make development and testing easier on me.
There are a bunch of great projects for running Kubernetes in development/lab environments, and others optimized for much larger enterprise environments, but I struggled to find a product that felt like a good fit for both in the way TCE was. My workloads are few and pretty simple so most enterprise K8s variants (Tanzu included) would feel like overkill, but I do need to ensure everything remains highly-available in the data centers at work.
So I set out to build my own! In the next couple of posts, I'll share the details of how I'm using Terraform to provision production-ready vanilla Kubernetes clusters on vSphere (complete with the vSphere Container Storage Interface plugin!) in a consistent and repeatable way. I also plan to document one of the ways I'm leveraging these clusters, which is using them as a part of a Gitlab CI/CD pipeline to churn out weekly VM template builds so I never again have to worry about my templates being out of date.
I've learned a ton in the process (and still have a lot more to learn), but today I'll start simply by describing how I'm leveraging Packer to create a single VM template ready to enter service as a Kubernetes compute node.
[HashiCorp Packer](https://www.packer.io/) is a free open-source tool designed to create consistent, repeatable machine images. It's pretty killer as a part of a CI/CD pipeline to kick off new builds based on a schedule or code commits, but also works great for creating builds on-demand. Packer uses the [HashiCorp Configuration Language (HCL)](https://developer.hashicorp.com/packer/docs/templates/hcl_templates) to describe all of the properties of a VM build in a concise and readable format.
You might ask why I would bother with using a powerful tool like Packer if I'm just going to be building a single template. Surely I could just do that by hand, right? And of course, you'd be right - but using an Infrastructure as Code tool even for one-off builds has some pretty big advantages.
- **It's fast.** Packer is able to build a complete VM (including pulling in all available OS and software updates) in just a few minutes, much faster than I could click through an installer on my own.
- **It's consistent.** Packer will follow the exact same steps for every build, removing the small variations (and typos!) that would surely show up if I did the builds manually.
- **It's great for testing changes.** Since Packer builds are so fast and consistent, it makes it incredibly easy to test changes as I go. I can be confident that the *only* changes between two builds will be the changes I deliberately introduced.
- **It's self-documenting.** The entire VM (and its guest OS) is described completely within the Packer HCL file(s), which I can review to remember which packages were installed, which user account(s) were created, what partition scheme was used, and anything else I might need to know.
- **It supports change tracking.** A Packer build is just a set of HCL files so it's easy to sync them with a version control system like Git to track (and revert) changes as needed.
Packer is also extremely versatile, and a broad set of [external plugins](https://developer.hashicorp.com/packer/plugins) expand its capabilities to support creating machines for basically any environment. For my needs, I'll be utilizing the [vsphere-iso](https://developer.hashicorp.com/packer/plugins/builders/vsphere/vsphere-iso) builder which uses the vSphere API to remotely build VMs directly on the hypervisors.
Sounds pretty cool, right? I'm not going to go too deep into "how to Packer" in this post, but HashiCorp does provide some [pretty good tutorials](https://developer.hashicorp.com/packer/tutorials) to help you get started.
Before being able to *use* Packer, you have to install it. On Debian/Ubuntu Linux, this process consists of adding the HashiCorp GPG key and software repository, and then simply installing the package:
You can learn how to install Packer on other systems by following [this tutorial from HashiCorp](https://developer.hashicorp.com/packer/tutorials/docker-get-started/get-started-install-cli).
Packer will need a user account with sufficient privileges in the vSphere environment to be able to create and manage a VM. I'd recommend using an account dedicated to automation tasks, and assigning it the required privileges listed in [the `vsphere-iso` documentation](https://developer.hashicorp.com/packer/plugins/builders/vsphere/vsphere-iso#required-vsphere-privileges).
My Kubernetes node template will use Ubuntu 20.04 LTS as the OS so I'll go ahead and download the [server installer ISO](https://releases.ubuntu.com/20.04.5/) and upload it to a vSphere datastore to make it available to Packer.
After the OS is installed and minimimally configured, I'll need to add in Kubernetes components like `containerd`, `kubectl`, `kubelet`, and `kubeadm`, and then apply a few additional tweaks to get it fully ready.
You can see the entirety of my Packer configuration [on GitHub](https://github.com/jbowdre/vsphere-k8s/tree/main/packer), but I'll talk through each file as we go along.
After quite a bit of experimentation, I've settled on a preferred way to organize my Packer build files. I've found that this structure makes the builds modular enough that it's easy to reuse components in other builds, but still consolidated enough to be easily manageable. This layout is, of course, largely subjective - it's just what works well *for me*:
```
.
├── certs
│ ├── ca.cer
├── data
│ ├── meta-data
│ └── user-data.pkrtpl.hcl
├── packer_cache
│ └── ssh_private_key_packer.pem
├── scripts
│ ├── cleanup-cloud-init.sh
│ ├── cleanup-subiquity.sh
│ ├── configure-sshd.sh
│ ├── disable-multipathd.sh
│ ├── disable-release-upgrade-motd.sh
│ ├── enable-vmware-customization.sh
│ ├── generalize.sh
│ ├── install-ca-certs.sh
│ ├── install-k8s.sh
│ ├── persist-cloud-init-net.sh
│ ├── update-packages.sh
│ ├── wait-for-cloud-init.sh
│ └── zero-disk.sh
├── ubuntu-k8s.auto.pkrvars.hcl
├── ubuntu-k8s.pkr.hcl
└── variables.pkr.hcl
```
- The `certs` folder holds the Base64-encoded PEM-formatted certificate of my [internal Certificate Authority](/ldaps-authentication-tanzu-community-edition/#prequisite) which will be automatically installed in the provisioned VM's trusted certificate store. I could also include additional root or intermediate certificates as needed, just as long as the file names end in `*.cer` (we'll see why later).
- The `data` folder stores files used for [generating the `cloud-init` configuration](#user-datapkrtplhcl) that will be used for the OS installation and configuration.
- The `scripts` directory holds a [collection of scripts](#post_install_scripts) used for post-install configuration tasks. Sure, I could just use a single large script, but using a bunch of smaller ones helps keep things modular and easy to reuse elsewhere.
-`variables.pkr.hcl` declares [all of the variables](#variablespkrhcl) which will be used in the Packer build, and sets the default values for some of them.
-`ubuntu-k8s.auto.pkrvars.hcl` [assigns values](#ubuntu-k8sautopkrvarshcl) to those variables. This is where most of the user-facing options will be configured, such as usernames, passwords, and environment settings.
-`ubuntu-k8s.pkr.hcl` is where the [build process](#ubuntu-k8spkrhcl) is actually described.
The first block in the file tells Packer about the minimum version requirements for Packer as well as the external plugins used for the build:
```
// BLOCK: packer
// The Packer configuration.
packer {
required_version = ">= 1.8.2"
required_plugins {
vsphere = {
version = ">= 1.0.8"
source = "github.com/hashicorp/vsphere"
}
sshkey = {
version = ">= 1.0.3"
source = "github.com/ivoronin/sshkey"
}
}
}
```
As I mentioned above, I'll be using the official [`vsphere` plugin](https://github.com/hashicorp/packer-plugin-vsphere) to handle the provisioning on my vSphere environment. I'll also make use of the [`sshkey` plugin](https://github.com/ivoronin/packer-plugin-sshkey) to easily handle the SSH keys.
#### `locals` block
Locals are a type of Packer variable which aren't explicitly declared in the `variables.pkr.hcl` file. They only exist within the context of a single build (hence the "local" name). Typical Packer variables are static and don't support string manipulation; locals, however, do support expressions that can be used to change their value on the fly. This makes them very useful when you need to combine variables (like a datastore name, path, filename) into a single string (such as in the highlighted line):
This block also makes use of the built-in [`templatefile()` function](https://developer.hashicorp.com/packer/docs/templates/hcl_templates/functions/file/templatefile) to insert build-specific variables into the `user-data` file for [`cloud-init`](https://cloud-init.io/) (more on that in a bit).
The `source` block tells the `vsphere-iso` builder how to connect to vSphere, what hardware specs to set on the VM, and what to do with the VM once the build has finished (convert it to template, export it to OVF, and so on).
This block brings everything together and executes the build. It calls the `source.vsphere-iso.ubuntu-k8s` block defined above, and also ties in a `file` and a few `shell` provisioners. `file` provisioners are used to copy files (like SSL CA certificates) into the VM, while the `shell` provisioners run commands and execute scripts. Those will be handy for the post-deployment configuration tasks, like updating and installing packages.
So you can see that the `ubuntu-k8s.pkr.hcl` file primarily focuses on the structure and form of the build, and it's written in such a way that it can be fairly easily adapted for building other types of VMs. Very few things in this file would have to be changed since so many of the properties are derived from the variables.
Before looking at the build-specific variable definitions, let's take a quick look at the variables I've told Packer that I intend to use. After all, Packer requires that variables be declared before they can be used.
Most of these carry descriptions with them so I won't restate them outside of the code block here:
```text
/*
DESCRIPTION:
Ubuntu Server 20.04 LTS variables using the Packer Builder for VMware vSphere (vsphere-iso).
Packer automatically knows to load variables defined in files ending in `*.auto.pkrvars.hcl`. Storing the variable values separately from the declarations in `variables.pkr.hcl` makes it easier to protect sensitive values.
So I'll start by telling Packer what credentials to use for connecting to vSphere, and what vSphere resources to deploy to:
```text
/*
DESCRIPTION:
Ubuntu Server 20.04 LTS Kubernetes node variables used by the Packer Plugin for VMware vSphere (vsphere-iso).
*/
// vSphere Credentials
vsphere_endpoint = "vcsa.lab.bowdre.net"
vsphere_username = "packer"
vsphere_password = "VMware1!"
vsphere_insecure_connection = true
// vSphere Settings
vsphere_datacenter = "NUC Site"
vsphere_cluster = "nuc-cluster"
vsphere_datastore = "nuchost-local"
vsphere_network = "MGT-Home 192.168.1.0"
vsphere_folder = "_Templates"
```
I'll then describe the properties of the VM itself:
```text
// Guest Operating System Settings
vm_guest_os_language = "en_US"
vm_guest_os_keyboard = "us"
vm_guest_os_timezone = "America/Chicago"
vm_guest_os_family = "linux"
vm_guest_os_type = "ubuntu64Guest"
// Virtual Machine Hardware Settings
vm_name = "k8s-u2004"
vm_firmware = "efi-secure"
vm_cdrom_type = "sata"
vm_cpu_count = 2
vm_cpu_cores = 1
vm_cpu_hot_add = true
vm_mem_size = 2048
vm_mem_hot_add = true
vm_disk_size = 30720
vm_disk_controller_type = ["pvscsi"]
vm_disk_thin_provisioned = true
vm_network_card = "vmxnet3"
common_vm_version = 19
common_tools_upgrade_policy = true
common_remove_cdrom = true
```
Then I'll configure Packer to convert the VM to a template once the build is finished:
```text
// Template and Content Library Settings
common_template_conversion = true
common_content_library_name = null
common_content_library_ovf = false
common_content_library_destroy = true
common_content_library_skip_export = true
// OVF Export Settings
common_ovf_export_enabled = false
common_ovf_export_overwrite = true
common_ovf_export_path = ""
```
Next, I'll tell it where to find the Ubuntu 20.04 ISO I downloaded and placed on a datastore, along with the SHA256 checksum to confirm its integrity:
And then I'll specify the VM's boot device order, as well as the boot command that will be used for loading the `cloud-init` coniguration into the Ubuntu installer:
Once the installer is booted and running, Packer will wait until the VM is available via SSH and then use these credentials to log in. (How will it be able to log in with those creds? We'll take a look at the `cloud-init` configuration in just a minute...)
Finally, I'll create two lists of scripts that will be run on the VM once the OS install is complete. The `post_install_scripts` will be run immediately after the operating system installation. The `update-packages.sh` script will cause a reboot, and then the set of `pre_final_scripts` will do some cleanup and prepare the VM to be converted to a template.
The last bit of this file also designates the desired version of Kubernetes to be installed.
Okay, so we've covered the Packer framework that creates the VM; now let's take a quick look at the `cloud-init` configuration that will allow the OS installation to proceed unattended.
See the bits that look `${ like_this }`? Those place-holders will take input from the [`locals` block of `ubuntu-k8s.pkr.hcl`](#locals-block) mentioned above. So that's how all the OS properties will get set, including the hostname, locale, LVM partition layout, username, password, and SSH key.
View the full file [here](https://github.com/jbowdre/vsphere-k8s/blob/main/packer/data/user-data.pkrtpl.hcl). (The `meta-data` file is [empty](https://github.com/jbowdre/vsphere-k8s/blob/main/packer/data/meta-data), by the way.)
After the OS install is completed, the `shell` provisioner will connect to the VM through SSH and run through some tasks. Remember how I keep talking about this build being modular? That goes down to the scripts, too, so I can use individual pieces in other builds without needing to do a lot of tweaking.
This simply holds up the process until the `/var/lib/cloud//instance/boot-finished` file has been created, signifying the completion of the `cloud-init` process:
```shell
#!/bin/bash -eu
echo '>> Waiting for cloud-init...'
while [ ! -f /var/lib/cloud/instance/boot-finished ]; do
sleep 1
done
```
#### `cleanup-subiquity.sh`
Next I clean up any network configs that may have been created during the install process:
```shell
#!/bin/bash -eu
if [ -f /etc/cloud/cloud.cfg.d/99-installer.cfg ]; then
sudo rm /etc/cloud/cloud.cfg.d/99-installer.cfg
echo 'Deleting subiquity cloud-init config'
fi
if [ -f /etc/cloud/cloud.cfg.d/subiquity-disable-cloudinit-networking.cfg ]; then
The [`file` provisioner](#build-block) mentioned above helpfully copied my custom CA certs to the `/tmp/certs/` folder on the VM; this script will install them into the certificate store:
I want to make sure that this VM keeps the same IP address following the reboot that will come in a few minutes, so I 'll set a quick `cloud-init` option to help make sure that happens:
```shell
#!/bin/sh -eu
echo '>> Preserving network settings...'
echo 'manual_cache_clean: True' | sudo tee -a /etc/cloud/cloud.cfg
```
#### `configure-sshd.sh`
Then I just set a few options for the `sshd` configuration, like disabling root login:
```shell
#!/bin/bash -eu
echo '>> Configuring SSH'
sudo sed -i 's/.*PermitRootLogin.*/PermitRootLogin no/' /etc/ssh/sshd_config
sudo sed -i 's/.*PubkeyAuthentication.*/PubkeyAuthentication yes/' /etc/ssh/sshd_config
sudo sed -i 's/.*PasswordAuthentication.*/PasswordAuthentication yes/' /etc/ssh/sshd_config
```
#### `install-k8s.sh`
This script is a little longer and takes care of all the Kubernetes-specific settings and packages that will need to be installed on the VM.
Next I'll install the Kubernetes components and (crucially) `apt-mark hold` them so they won't be automatically upgraded without it being a coordinated change:
Lastly, I'll be sure to update all installed packages (excepting the Kubernetes ones, of course), and then perform a reboot to make sure that any new kernel modules get loaded:
```shell
#!/bin/bash -eu
echo '>> Checking for and installing updates...'
sudo apt-get update && sudo apt-get -y upgrade
echo '>> Rebooting!'
sudo reboot
```
### `pre_final_scripts`
After the reboot, all that's left are some cleanup tasks to get the VM ready to be converted to a template and subsequently cloned and customized.
#### `cleanup-cloud-init.sh`
I'll start with cleaning up the `cloud-init` state:
```shell
#!/bin/bash -eu
echo '>> Cleaning up cloud-init state...'
sudo cloud-init clean -l
```
#### `enable-vmware-customization.sh`
And then be (re)enable the ability for VMware to be able to customize the guest successfully:
echo 'disable_vmware_customization: true' | sudo tee -a /etc/cloud/cloud.cfg
sudo vmware-toolbox-cmd config set deployPkg enable-custom-scripts true
```
#### `zero-disk.sh`
I'll also execute this handy script to free up unused space on the virtual disk. It works by creating a file which completely fills up the disk, and then deleting that file:
Now that all the ducks are nicely lined up, let's give them some marching orders and see what happens. All I have to do is open a terminal session to the folder containing the `.pkr.hcl` files, and then run the Packer build command:
The `-on-error=abort` option makes sure that the build will abort if any steps in the build fail, and `-force` tells Packer to delete any existing VMs/templates with the same name as the one I'm attempting to build.
And I can pop over to vSphere to confirm that everything looks right:
## Next steps
My brand new `k8s-u2004` template is ready for use! In the next post, I'll walk through the process of *manually* cloning this template to create my Kubernetes nodes, initializing the cluster, and installing the vSphere integrations. After that process is sorted out nicely, we'll take a look at how to use Terraform to do it all automagically. Stay tuned!
