exhibition-vm-controller: Snapshot-Revert as a Conservation Strategy

I work with a lot of digital art that was built for an OS nobody supports anymore. Flash and Pulse 3D on Windows XP. Browser-based works that depend on Java applets in Internet Explorer 8. The artworks themselves are intact. What breaks is the system underneath, and you cannot port that system without changing the art.
At ZKM the conservation strategy for this kind of work comes down to one rule: keep the whole operating system frozen and revert it on any fault, instead of trying to keep the software working. exhibition-vm-controller is the framework that runs this autonomously, at exhibition level.
The OS as the unit of recovery
A typical “auto-restart on crash” system watches a single process and respawns it. That works for software written in this decade. It does not work for software whose author is dead and whose runtime no longer exists. When the application has crashed, the OS underneath is usually also in a broken state that nobody knows how to clean up. Restarting the binary starts from a worse state than five minutes ago, and the damage adds up over the run of an exhibition.
So the framework treats the VM snapshot as the unit of recovery. You capture a “ready” snapshot of the artwork’s environment: Windows XP with everything launched, the artwork running, the splash screens dismissed. On any failure trigger the VM reverts. The OS is back in the exact “ready” state every time, with no drift, no leftover state and no manual intervention.
The triggers
The hard part is knowing when to revert. The framework watches for several things in parallel:
- Heartbeat timeout: the guest sends a heartbeat every second. After 10 s without a signal (default) the host reverts.
- VM shutdown: manual or unexpected, both are treated the same.
- QEMU guest agent unresponsive: the guest is alive at the hypervisor level, but its agent hangs.
- Application-specific error signals: the artwork itself can report an error through a guest-side script.
Heartbeats are platform-specific because the guest systems are. AutoIT scripts on Windows XP, Bash on Linux, AppleScript on Mac OS. Each of them sends one HTTP request to the host every second. When those requests stop, everything else starts.
The architecture
Host controller is a Python FastAPI service. It receives heartbeats, monitors VM state, controls the libvirt lifecycle (snapshot revert, restart) and offers a REST API so external monitoring can see what is going on.
Guest components are kept minimal: the QEMU guest agent where available, the heartbeat script in whatever the guest understands natively, an idle monitor for the application and a process watchdog. Anything more elaborate inside the guest breaks easily.
Display in production runs Openbox + virt-viewer --kiosk for full-screen exhibition mode. virt-manager is available during installation and maintenance.
Why VMs instead of emulation
Emulators are good for research. For an exhibition they are slower, the input timing is slightly wrong, and they often look a little off in ways visitors notice without being able to say why. A KVM/QEMU VM running the original OS on modern hardware is faster, faithful and cheaper to keep running for ten years. KVM has been stable for two decades. I am willing to bet it is still around in another twenty years.
Open source
Released under MIT, with a Zenodo DOI: 10.5281/zenodo.18652760. It is used alongside several historical works in the ZKM collection.
If you conserve software-based art and wonder whether VM snapshots work for autonomous exhibition operation: they do. We run it on real shows, and the code is open for you to use.