fedora-vm-builder

Ansible + OpenTofu builder for disposable Fedora VMs with kdump and kernel debuginfo pre-configured. Uses libvirt/KVM.

Prerequisites

• Linux host with libvirt/KVM (Fedora, RHEL, or CentOS Stream)
• Ansible (pip install ansible)
• OpenTofu >= 1.6 (tofu in PATH)
• sshpass (for password-based SSH to the VM)

Quick Start

# Install Ansible dependencies
ansible-galaxy collection install -r requirements.yml

# Create a VM (default: Fedora 44, kdump enabled, debuginfo installed)
ansible-playbook 01-create-vm.yml

# SSH into the VM
~/vm-lab/work/fedora01_login.sh

# Capture a vmcore (inside the VM)
echo c | sudo tee /proc/sysrq-trigger

# After the VM reboots, retrieve the vmcore
scp fedora@<vm-ip>:/var/crash/*/vmcore .
scp fedora@<vm-ip>:/usr/lib/debug/lib/modules/$(ssh fedora@<vm-ip> uname -r)/vmlinux .

# Destroy everything
ansible-playbook 99-destroy-all.yml

Configuration

Edit vars.yml before running. Key settings:

Variable	Default	Description
vm_prefix	fedora01	Name prefix for libvirt resources
vm_os_image	Fedora 44 Cloud Base	Fedora cloud image URL (38–44 selectable)
vm_vcpu	2	vCPUs
vm_memory	4	RAM in GB
vm_disk_size	20	Disk in GB
vm_crashkernel	512M	crashkernel reservation
vm_install_debuginfo	true	Install kernel debuginfo at build time

vars.yml is the only file you need to touch. Ansible converts it into terraform.tfvars.json (via the mapping in tfvars.yml) and hands it to the OpenTofu project in terraform/.

Layout

01-create-vm.yml     Create the VM and verify kdump is operational
99-destroy-all.yml   Tear everything down
vars.yml             User-facing settings
tfvars.yml           vars.yml -> OpenTofu input variable mapping
terraform/           Static OpenTofu project (no templating)
  versions.tf          OpenTofu + provider version constraints
  variables.tf         Typed, validated input variables
  main.tf              Storage pool + NAT network
  vm.tf                cloud-init disk, volumes, domain
  outputs.tf           vm_ip, vm_name, pool/network names
  templates/           cloud-init and network-config templates

The playbooks copy terraform/ into vm_tf_dir (~/vm-lab/work by default) and run OpenTofu there, so state and downloaded providers stay out of the repo and 99-destroy-all.yml can remove the whole lab directory.

What Gets Built

• libvirt storage pool (vm_prefix)
• NAT network with DHCP (vm_prefix_network)
• Fedora Cloud VM with:
  • User fedora / password fedora (sudo NOPASSWD)
  • kdump enabled with crashkernel=512M
  • kernel-debuginfo installed (vmlinux available)
  • Serial console + VNC

Coming from the Terraform version

Some releases drove Terraform instead of OpenTofu. If you have a lab left over from one, vm_tf_dir still holds a lock file pinned to registry.terraform.io, which tofu init will not install. You do not need to clean it up by hand — run

ansible-playbook 99-destroy-all.yml

first. The playbook re-initialises the directory before destroying, so OpenTofu takes over the existing state, and the virsh fallbacks catch anything it cannot. Then create a fresh VM as usual.

The terraform/ directory name and terraform.tfvars.json are kept as-is: OpenTofu reads both natively, and renaming them would churn every path in the docs and CI for no functional gain. versions.tf likewise keeps its terraform {} block, which OpenTofu accepts.

To go back to Terraform, set vm_tf_binary: terraform in vars.yml and regenerate the lock file — cloud.terraform.terraform drives either binary.

Provider version

The libvirt provider is pinned to ~> 0.8.3. The 0.9 line is a full schema rewrite in which libvirt_domain drops cloudinit, wait_for_lease and base_volume_id in favour of a raw libvirt-XML mapping; migrating is a separate piece of work.

License

MIT