Apache Calcite

This article summarizes my findings when I first begin to read Calcite’s codebase. The codebase is written in Java and is relatively well written in terms of software engineering practices.

SQL Query & Operators

Under org.apache.calcite.sql package, we can find the defitions related to SQL query and the oprators used in it. In Calcite, each SQL query is represented as a tree of operators (after being parsed by the org.apache.calcite.sql.parser module).

Each tree node (including the root node) is an SqlNode. SqlNode is an abstract class, of which using unparse method could help to convert an operator tree back to an SQL query in its string format. There are 3 types of SqlNodes:

• SqlOperator: a type of node in an operator tree (but not a node in the tree because SqlOperator is also an abstract class).
  • In practice, we would use SqlCall rather than SqlOperator though;
  • SqlKind is an enum that further classifies it into: QUERY, DML and DDL.
• SqlLiteral: a constant value (which means immutable), like NULL, boolean values, numbers and date & time values.
• SqlIdentifier: the identifier (or the name) of a database, table, column, etc. It could be compound though.

Query Optimizer

All related code should be udner org.apache.calcite.plan package. The two built-in optimizers are defined at org.apache.calcite.plan.hep and org.apache.calcite.plan.volcano respectively. They will use the various transformation rules defined in org.apache.calcite.rel.rules package.