Given a guarantee of snapshot-level consistency on a per-disk basis, are there architectural or operational risks that would render disk cloning an unsuitable method for provisioning Docker Swarm nodes?
How do you plan on creating a consistent snapshot across all manager nodes, or at least the majority of the manager nodes?
A disk-snapshot can only be used to recover the node the snapshot was taken from, as it will include the unique node id and must have the same ip when restored, in order to remain part of the swarm. If you restore more than half of the manager nodes and things don’t work as expected, the whole cluster would be headless and not able to do anything appart from running containers that are already running.
The intended approach is to stop a manager node, create a backup of the swarm metadata folder, then bring it back online so the replication log can catch up to the latest state in the cluster. This will require at least 3 manager nodes to work. Here are the details: https://docs.docker.com/engine/swarm/admin_guide/#back-up-the-swarm