HerdDB is a distributed Database, data is distributed among a cluster of server without the need of a shared storage.
HerdDB primary language is SQL and clients are encouraged to use both the JDBC Driver API and the low level API.
HerdDB is embeddable in any Java Virtual Machine, each node will access directly to local data without the use of the network.
HerdDB replication functions are built upon Apache ZooKeeper and Apache BookKeeper.
HerdDB is internally very similar to a NoSQL database and, basically, it is a key-value DB with an SQL abstraction layer which enables every user to leverage existing known-how and to port existing applications.
HerdDB has been designed for fast "writes" and for primary key read/update data access patterns.
HerdDB supports transactions and "committed read" isolation level
HerdDB uses Apache Calcite as SQL parser and SQL Planner
Data, as in any SQL database, is organized in tables and tables are grouped inside Tablespaces.
A Tablespace is the fundamental architectural brick upon which the replication is built and some DB features are available only among tables of the same tablespace:
- transactions may span only tables of the same tablespace
- subqueries may span only tables of the same tablespace
Replication is configured at tablespace level, so for each tablespace only one server is designed to be the 'leader' (manager) at a specific point in time, then you may configure a set of 'replicas'. The system automatically replicates data between replicas and handles server failures transparently.
Intoducing HerdDB - Pulsar Summit 2020 - Enrico Olivelli
Join the mailing list
HerdDB is under Apache 2 license.