How can I version my knowledge database?

I have a large number of documents (>= 1 million or more). I need to implement snapshot-style versioning, but I can't figure it out. It should be able to take snapshots from anywhere. (like git?) For example, if have Snap1, add data, take Snap2, and then roll back to Snap1, add data, and take Snap3. And it should only be able to query on each snapshot.

Snap1 ---- Snap2
      ---- Snap3 ---- Snap 4

How can I design such a structure? I was thinking of creating 1 million snapshot id - document ids every time a snapshot is created and then joining them, but that doesn't seem like a good idea. In more detail, this should be possible to implement in milvus (of course, snapshot metadata would probably have to be stored in RDBMS/NoSQL).