Since Holocron stores everything in object storage that can be geo-replicated a node truly becomes virtual. In the future replacing a node does not require bootstrapping a new node. Simply pointing the new compute at the crashed nodes S3 bucket+prefix will allow it to resume immediately.
Holocron is inspired by a blog post by Gunnar Morling discussing how distributed locks and leader elections are now possible in Amazon AWS S3 using conditional writes and the If-None-Match header support recently added to S3.
Holocron is highly experimental and only an hour-ish of hacking. Basically all it does at the moment is compete on an epoch lock file. It doesn't implement lock expiration or anything that makes it a complete implementation yet.