SQL Data Types

Arroyo currently supports a limited set of primitive types, shown in the table below. Because every Arroyo pipeline is a compiled Rust crate, each type has a corresponding Rust type.

Arroyo leverages Apache DataFusion for SQL support. DataFusion is in turn a part of the Apache Arrow project, a cross-language data format for columnar in-memory computation.

Depending on which part of the system you’re working with one or more of these will be relevant. For basic SQL users you should only need to deal with the SQL types.

These types can also be inserted as literals in sql. For types like INT vs BIGINT it will infer the type based on context.

Arroyo	Rust	Python	Arrow	Sql Types	Example Literals
Boolean	bool	bool	Boolean	BOOLEAN	TRUE, FALSE
Int32	i32	int	Int32	INT, INTEGER	0, 1, -2
Int64	i64	int	Int64	BIGINT	0, 1, -2
Uint32	u32	int	UInt32	INT UNSIGNED, INTEGER UNSIGNED	0, 1, 2
Uint64	u64	int	UInt64	BIGINT UNSIGNED	0, 1, 2
Float32	f32	float	Float32	FLOAT, REAL	0.0, -2.4, 1E-3
Float64	f64	float	Float64	DOUBLE	0.0, -2.4, 1E-35
Decimal128	i128	—	Decimal128	DECIMAL(precision, scale)	123.45, 0.001
String	String	str	Utf8	VARCHAR, CHAR, TEXT, STRING	"hello", "world"
Timestamp	SystemTime	—	Timestamp	TIMESTAMP	'2020-01-01', '2023-05-17T22:16:00.648662+00:00'
Bytes	Vec<u8>	—	Binary	BYTEA	X'A123' (hex)

Composite types

In addition to the primitive types above, Arroyo SQL supports two forms of composite types: arrays and structs.

Array types

Arrays group together zero or more elements of the same type. Arrays are declared by suffixing another type with [], for example an INT array can be declared with INT[]. Arrays values can be indexed using 1-indexed subscript notation (v[1] is the first element of v).

Arrays can be constructed via [] literals, like [1, 2, 3].

Arroyo provides a set of array functions for manipulating array values; arrays may also be unnested using the UNNEST operator.

Struct types

Structs allow combining related fields into a single value. Structs are declared using this syntax: struct<field_name field_type, ..>, and may contain any other type, including arrays and other structs.

For example:

CREATE TABLE events (
    properties STRUCT <
      user_id TEXT,
      timings INT[],
      name STRUCT <
        first TEXT,
        last TEXT
      >
    >
)

Structs fields can be accessed via . notation, for example properties.name.first.