Terraform Runner Shut Down the VM It Was Running On Mid-Apply

On July 4, 2026, a DevOps engineer discovered that Atlantis, a Terraform automation tool, destroyed its own runtime environment while applying infrastructure changes. Atlantis was deployed on a k3s virtual machine that was itself managed by the same Terraform stack, creating a circular dependency where the runner controlled the infrastructure it depended on. When a pull request modified the cpu.units attribute on one of the k3s VMs, the bpg/proxmox provider issued a full VM shutdown to apply the change, killing the Atlantis pod mid-execution and losing the apply state. The issue went undetected during planning because Terraform's plan phase does not account for provider-level behaviors like shutdown-before-update, making the hazard invisible until runtime. The problem was resolved by migrating Atlantis from the k3s deployment to a dedicated LXC container that is not managed by the same Terraform stack, breaking the circular dependency entirely.