This patch introduces an extensible data types API for HBase. It is inspired by the following systems:
- PostgreSQL. Postgres has a user-extensible data type API, which has been used to great effect by it's user community (ie, PostGIS). The desire is for HBase to expose an equally extensible data type API. One aspect of the Postgres data type is the ability to provide equivalence functions for index operations. This appears to be of critical performance utility for its execution engine.
- Orderly. Orderly handles the issue of compound rowkeys by providing convenience classes for handling these kinds of data types. This influence is reflected in the Struct and Union family of classes.
- Phoenix. The PDataType enum used in Phoenix provides type hints, similar Postgres's equivalence functions. These appear to be used during query execution for numerical type promotion.
This patch introduces an interface, DataType, along with a number of data type implementations based on the Bytes encoding. Also included are Struct and Union types, demonstrating simple implementations of compound types. Helper classes around the Struct implementation are also provided.
This patch does not address the type compatibility concerns expressed by Phoenix's PDataType API (ie, isComparableTo, isCoercibleTo); these will be addressed in
HBASE-8863.
This patch also provides DataType implementations based on the OrderedBytes encoding from
HBASE-8201.
Blocker HBASE-8201 OrderedBytes: an ordered encoding strategy
HIVE-6150 Take advantage of Native HBase Compound keys