Dolt is a single-node and embedded DBMS that incorporates Git-style versioning as a first-class entity. Dolt behaves like Git where it is a content addressable local database where the main objects are tables instead of files. In Dolt, a user creates a repository locally. The repository contains tables that can be read and updated using SQL. Similar to Git, writes are staged until the user issues a commit. Upon commit, the writes are appended to permanent storage.
Branch/merge semantics are supported allowing for the tables to evolve at a different pace for multiple users. This allows for loose collaboration on data as well as multiple views on the same core data. Merge conflicts are detected for schema and data conflicts. Data conflicts are cell-based, not line-based. Remote repositories allow for cooperation among repository instances. Clone, push, and pull semantics are all available.
Liquidata also created Dolthub, a website to host Dolt projects similar to GitHub.
Dolt stores tables in the N-ary Storage Model with clustered primary keys. The entire dataset is content-addressed as a Merkle Tree of component blocks. A Merkle tree is a hash-based data structure that is a generalization of the hash list. It is a tree structure in which each leaf node is a hash of a block of data, and each non-leaf node is a hash of its children. The boundaries for internal and leaf nodes are chosen by a rolling hash of the block contents.
Dolt does not support mutable database files, hence it does not explicitly take checkpoints. Dolt has a manifest that stores pointers to all currently active table files, which is updated atomically on every mutation of the database.
Dolt is not distributed at a system level. Dolt is designed to distribute the same database to multiple locations where it can be worked on in isolation and any edits from one location can be explicitly pulled or pushed to another location.
Dolt uses Snappy an open-sourced compression library that prioritizes speed over size. All chunks are compressed with Snappy before storage and are decompressed as they are read into the block cache. It is necessary to decompress data in order to process queries.
Dolt stores its dataset as a Merkle Tree of component blocks. The content-addressed blocks are stored in write-once table files with a static binary-searchable index at the end. When the table files grow to a number beyond a certain threshold, a compaction phase is run. New data is only written once. It is written to new table files containing the new chunks. These table files are flushed to disk before the manifest referencing them is updated.