Storage and transaction in mvSQLite
mvSQLite is the open-source, SQLite-compatible distributed database built on FoundationDB. You might be interested in the introduction post if this is the first time you heard of mvSQLite.
Immutable versions and branching are well-known concepts in Git-based modern software workflows, and mvSQLite brings these to the database world. mvSQLite keeps all historic versions of the database accessible. Different from traditional mechanisms for implementing Point-in-Time Recovery (PITR), mvSQLite does not checkout a historic version by replaying the logs - instead, the time complexity of reading from a time in the past is O(log n), identical to that of reading the current version.
Concurrency is a core property users expect from distributed databases. Uniquely in the distributed SQLite world, mvSQLite provides both read and write concurrency across a cluster of machines: readers do not conflict with each other, readers do not conflict with writers, and multiple writers sometimes conflict with each other.
Vanilla SQLite stores data in a single file on the local disk. From the perspective of SQLite's virtual filesystem (VFS), the minimal unit of data is a page - a byte sequence with a fixed length ranging from 512 to 65536 bytes.
