[VDBUH2023] - Franck Pachot - Anatomy of a distributed SQL database (YugabyteDB)

Porting all the features of PostgreSQL to a distributed database that “scales” horizontally is a challenge. But also the opportunity to modernize the underlying technologies of the DB, which becomes “cloud-native”: consensus protocols, logical clocks, automatic sharding. And to replace B-Tree indexes with LSM Tree and SSTables, more suited to SSD and distributed storage. YugabyteDB is open-source, and we will go into the details of the architecture, at the crossroads of PostgreSQL, Spanner, Cassandra, RocksDB… to better understand the reasons for a new database, and its underlying technology. We will discuss the advantages and the challenges of this unique architecture design: re-using the PostgreSQL query layer, plugged on top of a distributed storage and transaction layer