When using Git to store documents distributed and decentralized it can be considered as a database.
How would the ACID properties and the CAP theorem correspond to git in this case?
I think one has to distinguish between a single repository and the whole network of repositories.
ACID - for a single repository:
- A would be ok, as there are commits
- C depends on the use case and thus is not relevant
- I is the big question,
- if one is using only one branch that should be fine
- if one is considering multiple branches I would be fine if a rebase (resp. merge) without conflicts is possible to a master branch (or even all other branches) but is not fine if a rebase would result in a conflict.
- D depends on your hard disk but should generally be fine
ACID - for a distributed setup:
- would be the same as above plus, that I has to be seen with respect to all other clones on the network of repositories
CAP:
- C
- for a single repository this would be ok
- looking at a distributed setup this would only be true, if every read operation is preceded by a pull.
- A this would always be true since the local copy is always available
- P if a pull is not possible than one has to decide whether to sacrifice A or C (as in the PACELC theorem). Also one can see if the majority of remote repositories is available and use some quorum approach.
(I understand the CAP-C more like the ACID-A, see https://dba.stackexchange.com/a/202125/147543)
Is this a valid interpretation of ACID and CAP as it is used in the database domain? Or is there already an ongoing discussion about Git with regard to ACID and CAP?
