Abstract: A system performs a distributed version control on a data lake. A change to the data lake can be performed within a single atomic commit that defines a set of changes. Commits are built on top of each other and describe incremental changes, which creates a sequence of atomic commits that defines a history of changes. The commits can be identified by hash values, and branches can be defined to refer to commits. A new commit updates a branch head reference to the new commit so that the branch head always refers to the latest version of a chain of commits in the branch. The system can create tags that define, for example, a release version and can merge source commits onto target commits including merging the head of one branch onto the head of another branch.