Automating Multi-Cloud Infrastructure: Building Docker and K3s-Ready VHD/QCOW2 Images with OpenTofu and Packer

Introduction: The Multi-Cloud Standard Continuity Challenge

In the modern enterprise landscape, cloud-native architectures demand agility, repeatability, and strict environmental consistency. Operating across multi-cloud environments (such as AWS, Azure, Google Cloud, or on-premise OpenStack and Proxmox) often introduces a critical friction point: image drift. Discrepancies in base operating system images, kernel versions, and pre-installed runtimes can lead to unpredictable behavior in production environments.

To mitigate this risk, forward-thinking platform engineering teams are moving away from imperative configurations and embracing Immutable Infrastructure. By baking essential runtimes like Docker and K3s (a highly available, lightweight Kubernetes distribution) directly into virtual machine images (VHD, QCOW2) before deployment, organizations can drastically reduce provisioning times and guarantee state consistency. This technical blog explores how to seamlessly integrate Packer and OpenTofu to construct an automated, provider-agnostic image generation pipeline.

The Core Stack: Why Packer and OpenTofu?

Before diving into the implementation details, it is vital to understand why this specific toolchain provides a robust foundation for modern DevOps pipelines.

• Packer: Developed by HashiCorp (and maintained actively under open-source standards), Packer is the industry standard for creating identical machine images for multiple platforms from a single source configuration. It excels at launching a base VM, executing provisioners, and exporting the final artifact in formats like QCOW2 (for KVM/Proxmox) or VHD (for Hyper-V/Azure).
• OpenTofu: As a community-driven, open-source fork of Terraform, OpenTofu handles the declarative lifecycle management of your actual cloud infrastructure. Once Packer outputs the golden image, OpenTofu picks up the artifact to deploy compute instances, networking, and security parameters automatically.

By decoupling the building of the image (Packer) from the deployment of the image (OpenTofu), you establish a clean separation of concerns within your GitOps workflows.

Step-by-Step Architecture: Designing the Pipeline

The workflow follows a structured sequence: definition, compilation, provisioning, artifact generation, and deployment. Below is an architectural overview of how these pieces fit together:

“Infrastructure as Code is not just about provisioning servers; it is about establishing a repeatable, verifiable software supply chain for your entire operating environment.”

1. Configuring the Packer Template

Packer uses the HashiCorp Configuration Language (HCL) to define the source image and provisioners. To support multi-cloud targets, we define a single template that can output both QCOW2 and VHD formats. Below is a conceptual representation of our Packer configuration:

packer {
  required_plugins {
    qemu = {
      version = ">= 1.0.0"
      source  = "[github.com/hashicorp/qemu](https://github.com/hashicorp/qemu)"
    }
  }
}

We define a source "qemu" "ubuntu" block to pull a clean, upstream Ubuntu LTS ISO. The configuration specifies disk size, memory allocation, and the standard boot commands required to automate the initial OS installation via cloud-init or kickstart files.

2. Automating Docker and K3s Provisioning

Once the base operating system is initialized, Packer utilizes shell provisioners to execute installation scripts. To ensure production stability, we explicitly pin the versions of the software we install.

Our provisioning script executes the following high-level operations:

1. System Optimization: Updating package repositories, upgrading critical security patches, and configuring kernel modules required for container networking (such as overlay and br_netfilter).
2. Docker Engine Installation: Installing the official Docker Community Edition package, configuring the daemon to use systemd as the cgroup driver, and enabling the service on boot.
3. K3s Pre-configuration: Downloading the K3s binary and setting up the installation environment. Since this image will serve as a template, we download the images and binary files but do not initialize the cluster token or specific node roles yet. This ensures that when OpenTofu provisions the VM, each instance generates a unique machine ID and cryptographic keys.

3. Standardizing and Exporting Post-Build Artifacts

After the provisioners run successfully, Packer runs a sanitization step. This removes temporary SSH keys, clears shell histories, resets machine IDs, and cleans package caches to minimize the final image size. Packer then compiles the virtual disk into the targeted formats:

• QCOW2: Optimized for QEMU/KVM environments, featuring copy-on-write capabilities.
• VHD: Fixed-size or dynamic formats ready for upload directly into Microsoft Azure Managed Disks or Hyper-V clusters.

Seamless Handoff to OpenTofu

Once Packer stores the compiled image in a centralized registry or local storage pool, OpenTofu takes over. Using declarative state management, OpenTofu scans for the newly created image version and maps it to instance resources.

For instance, an OpenTofu module might define a cluster topology where a main.tf file references the custom QCOW2 image to spin up three control-plane nodes and five worker nodes dynamically. Because Docker and K3s binaries are already local to the disk, the cluster boots up and registers in seconds, completely avoiding long, error-prone initialization scripts during boot time.

Business and Operational Benefits

Implementing an automated image factory using OpenTofu and Packer yields substantial business advantages:

• Reduced Provisioning Latency: Traditional VM bootstrap scripts can take 10 to 15 minutes per instance to install large dependency trees. Pre-baked images reduce this time to under 60 seconds, drastically improving Auto-Scaling responsiveness during traffic spikes.
• Enhanced Security Posture: By baking CVE patches and hardened configuration baselines directly into the golden image, security teams can audit the infrastructure artifact before it is ever allowed to run in production.
• True Vendor Agility: Operating with standard formats like VHD and QCOW2 ensures that your application platform remains completely portable. If cloud providers alter their pricing metrics, shifting workloads to an alternative cloud or private hypervisor requires minimal architectural refactoring.

Conclusion and Next Steps

Automating the creation of pre-configured cloud images is a foundational pillar of modern platform engineering. By leveraging the combined strengths of Packer and OpenTofu, your organization can successfully eliminate environmental drift, optimize multi-cloud workflows, and accelerate delivery timelines. Begin by auditing your current manual configuration steps, translating them into declarative shell scripts, and incorporating image creation directly into your CI/CD pipelines.