The key-value conclusion #3
Description
I'm interested in the key-value conclusion. If the data is highly structured, e.g. tabular or hierarchical, then I imagine that a key might be "row x column y", or for (e.g. JSON) document data, a key might be "person.addresses.home.street_number". This could potentially make the storage quite inefficient/expensive.
I think that as long as
- changesets can refer to the smallest atoms of data,
- they can be reliably applied in forward/reverse direction as required
then storing the master version in (say) CSV or JSON format would be just as useful.
I also wonder whether mentioning the "Command/Query Responsibility Segregation" (CQRS) and Event Sourcing patterns in the Review of Existing Tools and Approaches might be useful.
https://en.wikipedia.org/wiki/Command%E2%80%93query_separation
https://martinfowler.com/eaaDev/EventSourcing.html
These can be useful when it is important to be able to undo or reconstruct sequences of operations.