Data Type Mapping

Values flow source native type → DataZen → target GetColumnTypeSQL. Azure SQL Database uses the SQL Server script builder (same mappings). Databricks and Oracle are not included in these tables. Self-mapping is also via DataZen, so a round-trip does not recreate the original native type.

Target cells are the type DataZen emits when creating or sinking a table, based on the DataZen type produced by the source driver. Sizes such as nvarchar(n) use the source column length when the driver reports it; max/LOB types use the engine’s unbounded mapping.

1. SQL Server as source

Azure SQL Database is identical. Native types listed from the LIVE covering table plus other types GetColumnTypeSQL accepts from DataZen.

SQL Server type	DataZen	SQL Server (via DataZen)	Fabric Warehouse	Snowflake	MySQL	Postgres
bigint	Int64	bigint	bigint	NUMBER(38, 0)	bigint	BIGINT
binary(n)	Byte[]	varbinary(n)	varbinary(n)	BINARY(67,108,864) 1	varbinary(n) / blob 1	BYTEA 1
bit	Boolean	bit	bit	BOOLEAN	tinyint 2	BOOLEAN
char(n)	String	nvarchar(n)	varchar(2n) 3	VARCHAR(134217728) 4	VARCHAR(n) / TEXT	TEXT 5
date	DateTime	datetime2	datetime2(6) 6	TIMESTAMP(3)	datetime	TIMESTAMP
datetime	DateTime	datetime2	datetime2(6) 6	TIMESTAMP(3)	datetime	TIMESTAMP
datetime2	DateTime	datetime2	datetime2(6) 6	TIMESTAMP(3)	datetime	TIMESTAMP
datetimeoffset	DateTimeOffset	datetimeoffset	datetime2(6) 7	TIMESTAMP_TZ(3)	datetime(6) 7	TIMESTAMPTZ
decimal(p,s)	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
float	Double	float	float	REAL	double	FLOAT
image	Byte[]	varbinary(max)	varbinary(max)	BINARY(67108864) 1	blob 1	BYTEA 1
int	Int32	int	int	NUMBER(38, 0)	int	INTEGER
money	Decimal	decimal(19,4) 8	decimal(19,4) 8	DECIMAL(19,4) 8	decimal(19,4) 8	decimal(19,4) 8
nchar(n)	String	nvarchar(n)	varchar(2n) 3	VARCHAR(134217728) 4	VARCHAR(n) / TEXT	TEXT 5
ntext	String	nvarchar(max)	varchar(max)	VARCHAR(134217728) 4	TEXT / LONGTEXT	TEXT
numeric(p,s)	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
nvarchar(n)	String	nvarchar(n)	varchar(2n) 3	VARCHAR(134217728) 4	VARCHAR(n) / TEXT	TEXT
nvarchar(max)	String	nvarchar(max)	varchar(max) 3	VARCHAR(134217728) 4	TEXT / LONGTEXT	TEXT
real	Single	float	real	REAL	double	REAL
rowversion / timestamp	Byte[]	varbinary(n) 9	varbinary(n) 9	BINARY(67108864) 1 9	varbinary(n) / blob 9	BYTEA 9
smalldatetime	DateTime	datetime2	datetime2(6) 6	TIMESTAMP(3)	datetime	TIMESTAMP
smallint	Int16	smallint	smallint	NUMBER(38, 0)	smallint	SMALLINT
smallmoney	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
text	String	nvarchar(max)	varchar(max)	VARCHAR(134217728) 4	TEXT / LONGTEXT	TEXT
time	TimeSpan	time(7)	time(6) 6	TIME(9) 10	time(6)	INTERVAL 10
tinyint	Byte	tinyint	smallint 11	NUMBER(3, 0)	smallint 11	SMALLINT
uniqueidentifier	Guid	uniqueidentifier	uniqueidentifier	VARCHAR	varchar(36)	TEXT
varbinary(n)	Byte[]	varbinary(n)	varbinary(n)	BINARY(67108864) 1	varbinary(n) / blob	BYTEA 1
varbinary(max)	Byte[]	varbinary(max)	varbinary(max)	BINARY(67108864) 1	blob	BYTEA
varchar(n)	String	nvarchar(n)	varchar(2n) 3	VARCHAR(134217728) 4	VARCHAR(n) / TEXT	TEXT
varchar(max)	String	nvarchar(max)	varchar(max)	VARCHAR(134217728) 4	TEXT / LONGTEXT	TEXT
xml	String	nvarchar(max)	varchar(max)	VARCHAR(134217728) 4	TEXT / LONGTEXT	TEXT

Binary MERGE/SINK literals: Snowflake wraps Base64 as TO_BINARY(..., 'BASE64') (NULL must not pass a format argument). Postgres uses decode(..., 'base64'). MySQL/Fabric emit quoted or X'hex' style literals depending on the builder.
MySQL has no BOOLEAN persist type here; bool maps to tinyint (0/1).
Fabric does not persist NVARCHAR. Character count is doubled into VARCHAR (UTF-8). varchar(max) is 16 MB per cell.
Snowflake VARCHAR ignores source length and uses the 128 MB maximum. Unquoted identifiers fold to UPPER CASE.
Postgres maps all strings to TEXT (PostgreSQL guidance: no performance benefit to VARCHAR(n)).
Fabric datetime2 and time are limited to 6 fractional digits (not 7).
Offset is dropped on Fabric (CAST(... AS DATETIME2(6))) and MySQL (UTC clock into DATETIME(6)).
SQL Server MONEY arrives as Decimal with precision 19 / scale 255; builders coerce scale to 4.
ROWVERSION/TIMESTAMP is not recreated as rowversion on any target; it becomes a byte array. SQL Server omits it from INSERT (engine-generated).
Time-of-day TimeSpan maps to TIME on SQL Server/Fabric/MySQL/Snowflake. Postgres maps TimeSpan to INTERVAL. Values ≥ 24 hours are not valid TIME.
DataZen byte is 0..255. Fabric has no TINYINT; MySQL signed TINYINT is −128..127, so DataZen uses SMALLINT.

2. Azure Fabric Warehouse as source

Persisted Warehouse types only. tinyint, nvarchar, datetime, datetimeoffset, money, xml, and image are not native Fabric types.

Fabric type	DataZen	SQL Server	Fabric (via DataZen)	Snowflake	MySQL	Postgres
bigint	Int64	bigint	bigint	NUMBER(38, 0)	bigint	BIGINT
binary(n) / varbinary(n)	Byte[]	varbinary(n)	varbinary(n)	BINARY(67108864) 1	varbinary(n) / blob	BYTEA
varbinary(max)	Byte[]	varbinary(max)	varbinary(max) 2	BINARY(67108864) 1	blob	BYTEA
bit	Boolean	bit	bit	BOOLEAN	tinyint	BOOLEAN
char(n)	String	nvarchar(n)	char(2n) / varchar	VARCHAR(134217728)	VARCHAR(n) / TEXT	TEXT
date	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
datetime2(6)	DateTime	datetime2	datetime2(6) 3	TIMESTAMP(3)	datetime	TIMESTAMP
decimal(p,s)	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
float	Double	float	float	REAL	double	FLOAT
int	Int32	int	int	NUMBER(38, 0)	int	INTEGER
real	Single	float	real	REAL	double	REAL
smallint	Int16	smallint	smallint	NUMBER(38, 0)	smallint	SMALLINT
time(6)	TimeSpan	time(7)	time(6) 3	TIME(9)	time(6)	INTERVAL
uniqueidentifier	Guid	uniqueidentifier	uniqueidentifier	VARCHAR	varchar(36)	TEXT
varchar(n)	String	nvarchar(n)	varchar(2n)	VARCHAR(134217728)	VARCHAR(n) / TEXT	TEXT
varchar(max)	String	nvarchar(max)	varchar(max) 2	VARCHAR(134217728)	TEXT / LONGTEXT	TEXT

Same binary literal rules as the SQL Server table (Base64 / decode / blob).
Warehouse VARCHAR(MAX) / VARBINARY(MAX) cap is 16 MB per cell.
Fractional seconds max is 6. Writing DateTimeOffset into Fabric always drops the offset.

3. Snowflake as source

Unquoted identifiers are stored UPPER CASE. NUMBER variants all travel as integer/decimal DataZen types; TIMESTAMP_LTZ/TZ become DateTimeOffset.

Snowflake type	DataZen	SQL Server	Fabric	Snowflake (via DataZen)	MySQL	Postgres
BINARY	Byte[]	varbinary(n/max)	varbinary(n/max)	BINARY(67108864) 1	blob / varbinary	BYTEA
BOOLEAN	Boolean	bit	bit	BOOLEAN	tinyint	BOOLEAN
CHAR(n)	String	nvarchar(n)	varchar(2n)	VARCHAR(134217728) 2	VARCHAR(n) / TEXT	TEXT
DATE	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
DECIMAL(p,s)	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
FLOAT	Double	float	float	REAL	double	FLOAT
NUMBER (integer)	Int32 / Int64 / Decimal	int / bigint / decimal	int / bigint / decimal	NUMBER(38, 0) or NUMBER(20, 0)	int / bigint	INTEGER / BIGINT / NUMERIC
REAL	Double / Single	float	float / real	REAL	double	FLOAT / REAL
TIME	TimeSpan / DateTime	time(7)	time(6)	TIME(9) 3	time(6)	INTERVAL / TIME
TIMESTAMP / TIMESTAMP_NTZ	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
TIMESTAMP_LTZ	DateTimeOffset	datetimeoffset	datetime2(6) 4	TIMESTAMP_TZ(3) 4	datetime(6) 4	TIMESTAMPTZ
TIMESTAMP_TZ	DateTimeOffset	datetimeoffset	datetime2(6) 4	TIMESTAMP_TZ(3)	datetime(6) 4	TIMESTAMPTZ
VARIANT	String	nvarchar(max)	varchar(max)	VARCHAR(134217728) 5	TEXT / LONGTEXT	TEXT
VARCHAR	String	nvarchar(n/max)	varchar(n/max)	VARCHAR(134217728) 2	VARCHAR / TEXT	TEXT

SINK MERGE must quote binary as Base64 inside TO_BINARY. TO_BINARY(NULL, 'BASE64') is invalid; NULL is emitted bare.
Length is not preserved. Table/column names must be UPPER CASE or quoted; temp @pipelinedata() objects need "@pipelinedata()".
TimeSpan is TIME(9), not TIMESTAMP. A clock string such as '00:00:01' is invalid as TIMESTAMP.
DateTimeOffset is formatted invariantly with offset (yyyy-MM-dd HH:mm:ss.fff zzz) and CAST to TIMESTAMP_TZ. Fabric/MySQL drop the zone.
VARIANT is not a DataZen type DataZen recreates; it is stored as text after the connector materializes it.

4. MySQL as source

Connector mappings from MySQLScriptBuilder comments. JSON is read as string. BIT(n>1) is typically Int64, not Boolean.

MySQL type	DataZen	SQL Server	Fabric	Snowflake	MySQL (via DataZen)	Postgres
bigint	Int64	bigint	bigint	NUMBER(38, 0)	bigint	BIGINT
binary / varbinary	Byte[]	varbinary(n)	varbinary(n)	BINARY(67108864)	varbinary(n) / blob	BYTEA
bit(1)	Boolean	bit	bit	BOOLEAN	tinyint	BOOLEAN
bit(n) n>1	Int64	bigint	bigint	NUMBER(38, 0)	bigint	BIGINT
blob / tinyblob / mediumblob / longblob	Byte[]	varbinary(max)	varbinary(max)	BINARY(67108864) 1	blob	BYTEA
char / varchar / text family	String	nvarchar(n/max)	varchar(n/max)	VARCHAR(134217728)	VARCHAR / TEXT / LONGTEXT 2	TEXT
date	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
datetime	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
decimal / numeric	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
double	Double	float	float	REAL	double	FLOAT
enum / set	String	nvarchar(n)	varchar(n)	VARCHAR(134217728)	VARCHAR(n) 3	TEXT
float	Single	float	real	REAL	double	REAL
int / integer	Int32	int	int	NUMBER(38, 0)	int	INTEGER
json	String	nvarchar(max)	varchar(max)	VARCHAR(134217728)	TEXT / LONGTEXT 4	TEXT
mediumint	Int32	int	int	NUMBER(38, 0)	int	INTEGER
smallint	Int16	smallint	smallint	NUMBER(38, 0)	smallint	SMALLINT
time	TimeSpan	time(7)	time(6)	TIME(9)	time(6)	INTERVAL
timestamp	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
tinyint (signed)	SByte	smallint	smallint	NUMBER(3, 0)	tinyint	SMALLINT
tinyint unsigned	Byte	tinyint	smallint	NUMBER(3, 0)	smallint	SMALLINT
year	Int16	smallint	smallint	NUMBER(38, 0)	smallint 5	SMALLINT

MySQL CAST AS LONGBLOB is not valid in seed/SELECT; blobs travel as Byte[] and are Base64-quoted for Snowflake.
String length: n < 2000 → VARCHAR(n); 2000..65535 → TEXT; larger or unknown → TEXT/LONGTEXT.
ENUM/SET are not recreated as ENUM/SET; they become ordinary strings.
The MySQL builder notes JSON as a source type, but GetColumnTypeSQL does not emit JSON — round-trip is TEXT.
YEAR is not recreated as YEAR.

5. PostgreSQL as source

Npgsql types. BIT(n) is BitArray (normalized to a 0/1 string for XML/SINK). INTERVAL is TimeSpan.

Postgres type	DataZen	SQL Server	Fabric	Snowflake	MySQL	Postgres (via DataZen)
bigint	Int64	bigint	bigint	NUMBER(38, 0)	bigint	BIGINT
bit(n)	BitArray	nvarchar / not native 1	varchar 1	VARCHAR 1	VARCHAR / TEXT 1	TEXT 1
boolean	Boolean	bit	bit	BOOLEAN	tinyint	BOOLEAN
bytea	Byte[]	varbinary(n/max)	varbinary(n/max)	BINARY(67108864)	blob	BYTEA 2
char / varchar / text	String	nvarchar(n/max)	varchar(n/max)	VARCHAR(134217728)	VARCHAR / TEXT	TEXT
date	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
double precision	Double	float	float	REAL	double	FLOAT
integer	Int32	int	int	NUMBER(38, 0)	int	INTEGER
interval	TimeSpan	time(7) 3	time(6) 3	TIME(9) 3	time(6) 3	INTERVAL
json / jsonb	String	nvarchar(max)	varchar(max)	VARCHAR(134217728)	TEXT	TEXT 4
money	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
numeric(p,s)	Decimal	decimal(p,s)	decimal(p,s)	DECIMAL(p,s)	decimal(p,s)	decimal(p,s)
real	Single	float	real	REAL	double	REAL
smallint	Int16	smallint	smallint	NUMBER(38, 0)	smallint	SMALLINT
time	TimeSpan	time(7)	time(6)	TIME(9)	time(6)	INTERVAL
timetz	DateTimeOffset / TimeSpan	datetimeoffset / time(7)	datetime2(6) / time(6)	TIMESTAMP_TZ(3) / TIME(9)	datetime(6) / time(6)	TIMESTAMPTZ / INTERVAL
timestamp	DateTime	datetime2	datetime2(6)	TIMESTAMP(3)	datetime	TIMESTAMP
timestamptz	DateTimeOffset	datetimeoffset	datetime2(6)	TIMESTAMP_TZ(3)	datetime(6)	TIMESTAMPTZ
uuid	Guid	uniqueidentifier	uniqueidentifier	VARCHAR	varchar(36)	TEXT 5
xml	String	nvarchar(max)	varchar(max)	VARCHAR(134217728)	TEXT	TEXT
xmin	UInt32 / String	bigint / nvarchar	bigint / varchar	NUMBER / VARCHAR	bigint / VARCHAR	NUMERIC / TEXT 6

BIT is kept native on the Postgres source table. BitArray is not XML-serializable; PostgresOperations normalizes cells to a 0/1 string, and SINK wraps as CAST('…' AS TEXT). GetColumnTypeSQL does not emit BIT.
Byte[] SINK uses decode('…', 'base64').
INTERVAL → TimeSpan → TIME on non-Postgres targets; day-long intervals will not fit TIME.
JSON/JSONB round-trip as TEXT, not json/jsonb.
UUID round-trip as TEXT, not uuid.
xmin is a system column (used in HWM/order tests). It is not created by GetColumnTypeSQL; include it only in SELECT lists.