Data Types

Flink SQL has a rich set of native data types available to users.

Data Type

A data type describes the logical type of a value in the table ecosystem. It can be used to declare input and/or output types of operations.

Flink’s data types are similar to the SQL standard’s data type terminology but also contain information about the nullability of a value for efficient handling of scalar expressions.

Examples of data types are:

  • INT
  • INT NOT NULL
  • INTERVAL DAY TO SECOND(3)
  • ROW<myField ARRAY<BOOLEAN>, myOtherField TIMESTAMP(3)>

A list of all pre-defined data types can be found below.

Data Types in the Table API

Java/Scala

Users of the JVM-based API work with instances of org.apache.flink.table.types.DataType within the Table API or when defining connectors, catalogs, or user-defined functions.

A DataType instance has two responsibilities:

  • Declaration of a logical type which does not imply a concrete physical representation for transmission or storage but defines the boundaries between JVM-based/Python languages and the table ecosystem.
  • Optional: Giving hints about the physical representation of data to the planner which is useful at the edges to other APIs.

For JVM-based languages, all pre-defined data types are available in org.apache.flink.table.api.DataTypes.

Python

Users of the Python API work with instances of pyflink.table.types.DataType within the Python Table API or when defining Python user-defined functions.

A DataType instance has such a responsibility:

  • Declaration of a logical type which does not imply a concrete physical representation for transmission or storage but defines the boundaries between Python languages and the table ecosystem.

For Python language, those types are available in pyflink.table.types.DataTypes.

Java

It is recommended to add a star import to your table programs for having a fluent API:

  1. import static org.apache.flink.table.api.DataTypes.*;
  2. DataType t = INTERVAL(DAY(), SECOND(3));

Scala

It is recommended to add a star import to your table programs for having a fluent API:

  1. import org.apache.flink.table.api.DataTypes._
  2. val t: DataType = INTERVAL(DAY(), SECOND(3))

Python

  1. from pyflink.table.types import DataTypes
  2. t = DataTypes.INTERVAL(DataTypes.DAY(), DataTypes.SECOND(3))

Physical Hints

Physical hints are required at the edges of the table ecosystem where the SQL-based type system ends and programming-specific data types are required. Hints indicate the data format that an implementation expects.

For example, a data source could express that it produces values for logical TIMESTAMPs using a java.sql.Timestamp class instead of using java.time.LocalDateTime which would be the default. With this information, the runtime is able to convert the produced class into its internal data format. In return, a data sink can declare the data format it consumes from the runtime.

Here are some examples of how to declare a bridging conversion class:

Java

  1. // tell the runtime to not produce or consume java.time.LocalDateTime instances
  2. // but java.sql.Timestamp
  3. DataType t = DataTypes.TIMESTAMP(3).bridgedTo(java.sql.Timestamp.class);
  4. // tell the runtime to not produce or consume boxed integer arrays
  5. // but primitive int arrays
  6. DataType t = DataTypes.ARRAY(DataTypes.INT().notNull()).bridgedTo(int[].class);

Scala

  1. // tell the runtime to not produce or consume java.time.LocalDateTime instances
  2. // but java.sql.Timestamp
  3. val t: DataType = DataTypes.TIMESTAMP(3).bridgedTo(classOf[java.sql.Timestamp])
  4. // tell the runtime to not produce or consume boxed integer arrays
  5. // but primitive int arrays
  6. val t: DataType = DataTypes.ARRAY(DataTypes.INT().notNull()).bridgedTo(classOf[Array[Int]])

Attention Please note that physical hints are usually only required if the API is extended. Users of predefined sources/sinks/functions do not need to define such hints. Hints within a table program (e.g. field.cast(TIMESTAMP(3).bridgedTo(Timestamp.class))) are ignored.

List of Data Types

This section lists all pre-defined data types.

Java/Scala

For the JVM-based Table API those types are also available in org.apache.flink.table.api.DataTypes.

Python

For the Python Table API, those types are available in pyflink.table.types.DataTypes.

The default planner supports the following set of SQL types:

Data TypeRemarks for Data Type
CHAR
VARCHAR
STRING
BOOLEAN
BINARY
VARBINARY
BYTES
DECIMALSupports fixed precision and scale.
TINYINT
SMALLINT
INTEGER
BIGINT
FLOAT
DOUBLE
DATE
TIMESupports only a precision of 0.
TIMESTAMP
TIMESTAMP_LTZ
INTERVALSupports only interval of MONTH and SECOND(3).
ARRAY
MULTISET
MAP
ROW
RAW
Structured typesOnly exposed in user-defined functions yet.

Character Strings

CHAR

Data type of a fixed-length character string.

Declaration

SQL

  1. CHAR
  2. CHAR(n)

Java/Scala

  1. DataTypes.CHAR(n)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.StringXXDefault
byte[]XXAssumes UTF-8 encoding.
org.apache.flink.table.data.StringDataXXInternal data structure.

Python

  1. Not supported.

The type can be declared using CHAR(n) where n is the number of code points. n must have a value between 1 and 2,147,483,647 (both inclusive). If no length is specified, n is equal to 1.

VARCHAR / STRING

Data type of a variable-length character string.

Declaration

SQL

  1. VARCHAR
  2. VARCHAR(n)
  3. STRING

Java/Scala

  1. DataTypes.VARCHAR(n)
  2. DataTypes.STRING()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.StringXXDefault
byte[]XXAssumes UTF-8 encoding.
org.apache.flink.table.data.StringDataXXInternal data structure.

Python

  1. DataTypes.VARCHAR(n)
  2. DataTypes.STRING()

Attention The specified maximum number of code points n in DataTypes.VARCHAR(n) must be 2,147,483,647 currently.

The type can be declared using VARCHAR(n) where n is the maximum number of code points. n must have a value between 1 and 2,147,483,647 (both inclusive). If no length is specified, n is equal to 1.

STRING is a synonym for VARCHAR(2147483647).

Binary Strings

BINARY

Data type of a fixed-length binary string (=a sequence of bytes).

Declaration

SQL

  1. BINARY
  2. BINARY(n)

Java/Scala

  1. DataTypes.BINARY(n)

Bridging to JVM Types

Java TypeInputOutputRemarks
byte[]XXDefault

Python

  1. Not supported.

The type can be declared using BINARY(n) where n is the number of bytes. n must have a value between 1 and 2,147,483,647 (both inclusive). If no length is specified, n is equal to 1.

VARBINARY / BYTES

Data type of a variable-length binary string (=a sequence of bytes).

Declaration

SQL

  1. VARBINARY
  2. VARBINARY(n)
  3. BYTES

Java/Scala

  1. DataTypes.VARBINARY(n)
  2. DataTypes.BYTES()

Bridging to JVM Types

Java TypeInputOutputRemarks
byte[]XXDefault

Python

  1. DataTypes.VARBINARY(n)
  2. DataTypes.BYTES()

Attention The specified maximum number of bytes n in DataTypes.VARBINARY(n) must be 2,147,483,647 currently.

The type can be declared using VARBINARY(n) where n is the maximum number of bytes. n must have a value between 1 and 2,147,483,647 (both inclusive). If no length is specified, n is equal to 1.

BYTES is a synonym for VARBINARY(2147483647).

Exact Numerics

DECIMAL

Data type of a decimal number with fixed precision and scale.

Declaration

SQL

  1. DECIMAL
  2. DECIMAL(p)
  3. DECIMAL(p, s)
  4. DEC
  5. DEC(p)
  6. DEC(p, s)
  7. NUMERIC
  8. NUMERIC(p)
  9. NUMERIC(p, s)

Java/Scala

  1. DataTypes.DECIMAL(p, s)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.math.BigDecimalXXDefault
org.apache.flink.table.data.DecimalDataXXInternal data structure.

Python

  1. DataTypes.DECIMAL(p, s)

Attention The precision and scale specified in DataTypes.DECIMAL(p, s) must be 38 and 18 separately currently.

The type can be declared using DECIMAL(p, s) where p is the number of digits in a number (precision) and s is the number of digits to the right of the decimal point in a number (scale). p must have a value between 1 and 38 (both inclusive). s must have a value between 0 and p (both inclusive). The default value for p is 10. The default value for s is 0.

NUMERIC(p, s) and DEC(p, s) are synonyms for this type.

TINYINT

Data type of a 1-byte signed integer with values from -128 to 127.

Declaration

SQL

  1. TINYINT

Java/Scala

  1. DataTypes.TINYINT()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.ByteXXDefault
byteX(X)Output only if type is not nullable.

Python

  1. DataTypes.TINYINT()

SMALLINT

Data type of a 2-byte signed integer with values from -32,768 to 32,767.

Declaration

SQL

  1. SMALLINT

Java/Scala

  1. DataTypes.SMALLINT()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.ShortXXDefault
shortX(X)Output only if type is not nullable.

Python

  1. DataTypes.SMALLINT()

INT

Data type of a 4-byte signed integer with values from -2,147,483,648 to 2,147,483,647.

Declaration

SQL

  1. INT
  2. INTEGER

Java/Scala

  1. DataTypes.INT()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.IntegerXXDefault
intX(X)Output only if type is not nullable.

Python

  1. DataTypes.INT()

INTEGER is a synonym for this type.

BIGINT

Data type of an 8-byte signed integer with values from -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807.

Declaration

SQL

  1. BIGINT

Java/Scala

  1. DataTypes.BIGINT()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.LongXXDefault
longX(X)Output only if type is not nullable.

Python

  1. DataTypes.BIGINT()

Approximate Numerics

FLOAT

Data type of a 4-byte single precision floating point number.

Compared to the SQL standard, the type does not take parameters.

Declaration

SQL

  1. FLOAT

Java/Scala

  1. DataTypes.FLOAT()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.FloatXXDefault
floatX(X)Output only if type is not nullable.

Python

  1. DataTypes.FLOAT()

DOUBLE

Data type of an 8-byte double precision floating point number.

Declaration

SQL

  1. DOUBLE
  2. DOUBLE PRECISION

Java/Scala

  1. DataTypes.DOUBLE()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.DoubleXXDefault
doubleX(X)Output only if type is not nullable.

Python

  1. DataTypes.DOUBLE()

DOUBLE PRECISION is a synonym for this type.

Date and Time

DATE

Data type of a date consisting of year-month-day with values ranging from 0000-01-01 to 9999-12-31.

Compared to the SQL standard, the range starts at year 0000.

Declaration

SQL

  1. DATE

Java/Scala

  1. DataTypes.DATE()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.LocalDateXXDefault
java.sql.DateXX
java.lang.IntegerXXDescribes the number of days since epoch.
intX(X)Describes the number of days since epoch.
Output only if type is not nullable.

Python

  1. DataTypes.DATE()

TIME

Data type of a time without time zone consisting of hour:minute:second[.fractional] with up to nanosecond precision and values ranging from 00:00:00.000000000 to 23:59:59.999999999.

SQL/Java/Scala

Compared to the SQL standard, leap seconds (23:59:60 and 23:59:61) are not supported as the semantics are closer to java.time.LocalTime. A time with time zone is not provided.

Python

Compared to the SQL standard, leap seconds (23:59:60 and 23:59:61) are not supported. A time with time zone is not provided.

Declaration

SQL

  1. TIME
  2. TIME(p)

Java/Scala

  1. DataTypes.TIME(p)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.LocalTimeXXDefault
java.sql.TimeXX
java.lang.IntegerXXDescribes the number of milliseconds of the day.
intX(X)Describes the number of milliseconds of the day.
Output only if type is not nullable.
java.lang.LongXXDescribes the number of nanoseconds of the day.
longX(X)Describes the number of nanoseconds of the day.
Output only if type is not nullable.

Python

  1. DataTypes.TIME(p)

Attention The precision specified in DataTypes.TIME(p) must be 0 currently.

The type can be declared using TIME(p) where p is the number of digits of fractional seconds (precision). p must have a value between 0 and 9 (both inclusive). If no precision is specified, p is equal to 0.

TIMESTAMP

Data type of a timestamp without time zone consisting of year-month-day hour:minute:second[.fractional] with up to nanosecond precision and values ranging from 0000-01-01 00:00:00.000000000 to 9999-12-31 23:59:59.999999999.

SQL/Java/Scala

Compared to the SQL standard, leap seconds (23:59:60 and 23:59:61) are not supported as the semantics are closer to java.time.LocalDateTime.

A conversion from and to BIGINT (a JVM long type) is not supported as this would imply a time zone. However, this type is time zone free. For more java.time.Instant-like semantics use TIMESTAMP_LTZ.

Python

Compared to the SQL standard, leap seconds (23:59:60 and 23:59:61) are not supported.

A conversion from and to BIGINT is not supported as this would imply a time zone. However, this type is time zone free. If you have such a requirement please use TIMESTAMP_LTZ.

Declaration

SQL

  1. TIMESTAMP
  2. TIMESTAMP(p)
  3. TIMESTAMP WITHOUT TIME ZONE
  4. TIMESTAMP(p) WITHOUT TIME ZONE

Java/Scala

  1. DataTypes.TIMESTAMP(p)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.LocalDateTimeXXDefault
java.sql.TimestampXX
org.apache.flink.table.data.TimestampDataXXInternal data structure.

Python

  1. DataTypes.TIMESTAMP(p)

Attention The precision specified in DataTypes.TIMESTAMP(p) must be 3 currently.

The type can be declared using TIMESTAMP(p) where p is the number of digits of fractional seconds (precision). p must have a value between 0 and 9 (both inclusive). If no precision is specified, p is equal to 6.

TIMESTAMP(p) WITHOUT TIME ZONE is a synonym for this type.

TIMESTAMP WITH TIME ZONE

Data type of a timestamp with time zone consisting of year-month-day hour:minute:second[.fractional] zone with up to nanosecond precision and values ranging from 0000-01-01 00:00:00.000000000 +14:59 to 9999-12-31 23:59:59.999999999 -14:59.

SQL/Java/Scala

Compared to the SQL standard, leap seconds (23:59:60 and 23:59:61) are not supported as the semantics are closer to java.time.OffsetDateTime.

Python

Compared to the SQL standard, leap seconds (23:59:60 and 23:59:61) are not supported.

Compared to TIMESTAMP_LTZ, the time zone offset information is physically stored in every datum. It is used individually for every computation, visualization, or communication to external systems.

Declaration

SQL

  1. TIMESTAMP WITH TIME ZONE
  2. TIMESTAMP(p) WITH TIME ZONE

Java/Scala

  1. DataTypes.TIMESTAMP_WITH_TIME_ZONE(p)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.OffsetDateTimeXXDefault
java.time.ZonedDateTimeXIgnores the zone ID.

Python

  1. Not supported.

SQL/Java/Scala

The type can be declared using TIMESTAMP(p) WITH TIME ZONE where p is the number of digits of fractional seconds (precision). p must have a value between 0 and 9 (both inclusive). If no precision is specified, p is equal to 6.

Python

TIMESTAMP_LTZ

Data type of a timestamp with local time zone consisting of year-month-day hour:minute:second[.fractional] zone with up to nanosecond precision and values ranging from 0000-01-01 00:00:00.000000000 +14:59 to 9999-12-31 23:59:59.999999999 -14:59.

SQL/Java/Scala

Leap seconds (23:59:60 and 23:59:61) are not supported as the semantics are closer to java.time.OffsetDateTime.

Compared to TIMESTAMP WITH TIME ZONE, the time zone offset information is not stored physically in every datum. Instead, the type assumes java.time.Instant semantics in UTC time zone at the edges of the table ecosystem. Every datum is interpreted in the local time zone configured in the current session for computation and visualization.

Python

Leap seconds (23:59:60 and 23:59:61) are not supported.

Compared to TIMESTAMP WITH TIME ZONE, the time zone offset information is not stored physically in every datum. Every datum is interpreted in the local time zone configured in the current session for computation and visualization.

This type fills the gap between time zone free and time zone mandatory timestamp types by allowing the interpretation of UTC timestamps according to the configured session time zone.

Declaration

SQL

  1. TIMESTAMP_LTZ
  2. TIMESTAMP_LTZ(p)
  3. TIMESTAMP WITH LOCAL TIME ZONE
  4. TIMESTAMP(p) WITH LOCAL TIME ZONE

Java/Scala

  1. DataTypes.TIMESTAMP_LTZ(p)
  2. DataTypes.TIMESTAMP_WITH_LOCAL_TIME_ZONE(p)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.InstantXXDefault
java.lang.IntegerXXDescribes the number of seconds since epoch.
intX(X)Describes the number of seconds since epoch.
Output only if type is not nullable.
java.lang.LongXXDescribes the number of milliseconds since epoch.
longX(X)Describes the number of milliseconds since epoch.
Output only if type is not nullable.
java.sql.TimestampXXDescribes the number of milliseconds since epoch.
org.apache.flink.table.data.TimestampDataXXInternal data structure.

Python

  1. DataTypes.TIMESTAMP_LTZ(p)
  2. DataTypes.TIMESTAMP_WITH_LOCAL_TIME_ZONE(p)

Attention The precision specified in DataTypes.TIMESTAMP_LTZ(p) must be 3 currently.

The type can be declared using TIMESTAMP_LTZ(p) where p is the number of digits of fractional seconds (precision). p must have a value between 0 and 9 (both inclusive). If no precision is specified, p is equal to 6.

TIMESTAMP(p) WITH LOCAL TIME ZONE is a synonym for this type.

INTERVAL YEAR TO MONTH

Data type for a group of year-month interval types.

The type must be parameterized to one of the following resolutions:

  • interval of years,
  • interval of years to months,
  • or interval of months.

An interval of year-month consists of +years-months with values ranging from -9999-11 to +9999-11.

The value representation is the same for all types of resolutions. For example, an interval of months of 50 is always represented in an interval-of-years-to-months format (with default year precision): +04-02.

Declaration

SQL

  1. INTERVAL YEAR
  2. INTERVAL YEAR(p)
  3. INTERVAL YEAR(p) TO MONTH
  4. INTERVAL MONTH

Java/Scala

  1. DataTypes.INTERVAL(DataTypes.YEAR())
  2. DataTypes.INTERVAL(DataTypes.YEAR(p))
  3. DataTypes.INTERVAL(DataTypes.YEAR(p), DataTypes.MONTH())
  4. DataTypes.INTERVAL(DataTypes.MONTH())

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.PeriodXXIgnores the days part. Default
java.lang.IntegerXXDescribes the number of months.
intX(X)Describes the number of months.
Output only if type is not nullable.

Python

  1. DataTypes.INTERVAL(DataTypes.YEAR())
  2. DataTypes.INTERVAL(DataTypes.YEAR(p))
  3. DataTypes.INTERVAL(DataTypes.YEAR(p), DataTypes.MONTH())
  4. DataTypes.INTERVAL(DataTypes.MONTH())

The type can be declared using the above combinations where p is the number of digits of years (year precision). p must have a value between 1 and 4 (both inclusive). If no year precision is specified, p is equal to 2.

INTERVAL DAY TO SECOND

Data type for a group of day-time interval types.

The type must be parameterized to one of the following resolutions with up to nanosecond precision:

  • interval of days,
  • interval of days to hours,
  • interval of days to minutes,
  • interval of days to seconds,
  • interval of hours,
  • interval of hours to minutes,
  • interval of hours to seconds,
  • interval of minutes,
  • interval of minutes to seconds,
  • or interval of seconds.

An interval of day-time consists of +days hours:months:seconds.fractional with values ranging from -999999 23:59:59.999999999 to +999999 23:59:59.999999999. The value representation is the same for all types of resolutions. For example, an interval of seconds of 70 is always represented in an interval-of-days-to-seconds format (with default precisions): +00 00:01:10.000000.

Declaration

SQL

  1. INTERVAL DAY
  2. INTERVAL DAY(p1)
  3. INTERVAL DAY(p1) TO HOUR
  4. INTERVAL DAY(p1) TO MINUTE
  5. INTERVAL DAY(p1) TO SECOND(p2)
  6. INTERVAL HOUR
  7. INTERVAL HOUR TO MINUTE
  8. INTERVAL HOUR TO SECOND(p2)
  9. INTERVAL MINUTE
  10. INTERVAL MINUTE TO SECOND(p2)
  11. INTERVAL SECOND
  12. INTERVAL SECOND(p2)

Java/Scala

  1. DataTypes.INTERVAL(DataTypes.DAY())
  2. DataTypes.INTERVAL(DataTypes.DAY(p1))
  3. DataTypes.INTERVAL(DataTypes.DAY(p1), DataTypes.HOUR())
  4. DataTypes.INTERVAL(DataTypes.DAY(p1), DataTypes.MINUTE())
  5. DataTypes.INTERVAL(DataTypes.DAY(p1), DataTypes.SECOND(p2))
  6. DataTypes.INTERVAL(DataTypes.HOUR())
  7. DataTypes.INTERVAL(DataTypes.HOUR(), DataTypes.MINUTE())
  8. DataTypes.INTERVAL(DataTypes.HOUR(), DataTypes.SECOND(p2))
  9. DataTypes.INTERVAL(DataTypes.MINUTE())
  10. DataTypes.INTERVAL(DataTypes.MINUTE(), DataTypes.SECOND(p2))
  11. DataTypes.INTERVAL(DataTypes.SECOND())
  12. DataTypes.INTERVAL(DataTypes.SECOND(p2))

Bridging to JVM Types

Java TypeInputOutputRemarks
java.time.DurationXXDefault
java.lang.LongXXDescribes the number of milliseconds.
longX(X)Describes the number of milliseconds.
Output only if type is not nullable.

Python

  1. DataTypes.INTERVAL(DataTypes.DAY())
  2. DataTypes.INTERVAL(DataTypes.DAY(p1))
  3. DataTypes.INTERVAL(DataTypes.DAY(p1), DataTypes.HOUR())
  4. DataTypes.INTERVAL(DataTypes.DAY(p1), DataTypes.MINUTE())
  5. DataTypes.INTERVAL(DataTypes.DAY(p1), DataTypes.SECOND(p2))
  6. DataTypes.INTERVAL(DataTypes.HOUR())
  7. DataTypes.INTERVAL(DataTypes.HOUR(), DataTypes.MINUTE())
  8. DataTypes.INTERVAL(DataTypes.HOUR(), DataTypes.SECOND(p2))
  9. DataTypes.INTERVAL(DataTypes.MINUTE())
  10. DataTypes.INTERVAL(DataTypes.MINUTE(), DataTypes.SECOND(p2))
  11. DataTypes.INTERVAL(DataTypes.SECOND())
  12. DataTypes.INTERVAL(DataTypes.SECOND(p2))

The type can be declared using the above combinations where p1 is the number of digits of days (day precision) and p2 is the number of digits of fractional seconds (fractional precision). p1 must have a value between 1 and 6 (both inclusive). p2 must have a value between 0 and 9 (both inclusive). If no p1 is specified, it is equal to 2 by default. If no p2 is specified, it is equal to 6 by default.

Constructured Data Types

ARRAY

Data type of an array of elements with same subtype.

Compared to the SQL standard, the maximum cardinality of an array cannot be specified but is fixed at 2,147,483,647. Also, any valid type is supported as a subtype.

Declaration

SQL

  1. ARRAY<t>
  2. t ARRAY

Java/Scala

  1. DataTypes.ARRAY(t)

Bridging to JVM Types

Java TypeInputOutputRemarks
t[](X)(X)Depends on the subtype. Default
java.util.List<t>XX
subclass of java.util.List<t>X
org.apache.flink.table.data.ArrayDataXXInternal data structure.

Python

  1. DataTypes.ARRAY(t)

The type can be declared using ARRAY<t> where t is the data type of the contained elements.

t ARRAY is a synonym for being closer to the SQL standard. For example, INT ARRAY is equivalent to ARRAY<INT>.

MAP

Data type of an associative array that maps keys (including NULL) to values (including NULL). A map cannot contain duplicate keys; each key can map to at most one value.

There is no restriction of element types; it is the responsibility of the user to ensure uniqueness.

The map type is an extension to the SQL standard.

Declaration

SQL

  1. MAP<kt, vt>

Java/Scala

  1. DataTypes.MAP(kt, vt)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.util.Map<kt, vt>XXDefault
subclass of java.util.Map<kt, vt>X
org.apache.flink.table.data.MapDataXXInternal data structure.

Python

  1. DataTypes.MAP(kt, vt)

The type can be declared using MAP<kt, vt> where kt is the data type of the key elements and vt is the data type of the value elements.

MULTISET

Data type of a multiset (=bag). Unlike a set, it allows for multiple instances for each of its elements with a common subtype. Each unique value (including NULL) is mapped to some multiplicity.

There is no restriction of element types; it is the responsibility of the user to ensure uniqueness.

Declaration

SQL

  1. MULTISET<t>
  2. t MULTISET

Java/Scala

  1. DataTypes.MULTISET(t)

Bridging to JVM Types

Java TypeInputOutputRemarks
java.util.Map<t, java.lang.Integer>XXAssigns each value to an integer multiplicity. Default
subclass of java.util.Map<t, java.lang.Integer>>X
org.apache.flink.table.data.MapDataXXInternal data structure.

Python

  1. DataTypes.MULTISET(t)

The type can be declared using MULTISET<t> where t is the data type of the contained elements.

t MULTISET is a synonym for being closer to the SQL standard. For example, INT MULTISET is equivalent to MULTISET<INT>.

ROW

Data type of a sequence of fields.

A field consists of a field name, field type, and an optional description. The most specific type of a row of a table is a row type. In this case, each column of the row corresponds to the field of the row type that has the same ordinal position as the column.

Compared to the SQL standard, an optional field description simplifies the handling with complex structures.

A row type is similar to the STRUCT type known from other non-standard-compliant frameworks.

Declaration

SQL

  1. ROW<n0 t0, n1 t1, ...>
  2. ROW<n0 t0 'd0', n1 t1 'd1', ...>
  3. ROW(n0 t0, n1 t1, ...)
  4. ROW(n0 t0 'd0', n1 t1 'd1', ...)

Java/Scala

  1. DataTypes.ROW(DataTypes.FIELD(n0, t0), DataTypes.FIELD(n1, t1), ...)
  2. DataTypes.ROW(DataTypes.FIELD(n0, t0, d0), DataTypes.FIELD(n1, t1, d1), ...)

Bridging to JVM Types

Java TypeInputOutputRemarks
org.apache.flink.types.RowXXDefault
org.apache.flink.table.data.RowDataXXInternal data structure.

Python

  1. DataTypes.ROW([DataTypes.FIELD(n0, t0), DataTypes.FIELD(n1, t1), ...])
  2. DataTypes.ROW([DataTypes.FIELD(n0, t0, d0), DataTypes.FIELD(n1, t1, d1), ...])

The type can be declared using ROW<n0 t0 'd0', n1 t1 'd1', ...> where n is the unique name of a field, t is the logical type of a field, d is the description of a field.

ROW(...) is a synonym for being closer to the SQL standard. For example, ROW(myField INT, myOtherField BOOLEAN) is equivalent to ROW<myField INT, myOtherField BOOLEAN>.

User-Defined Data Types

Java/Scala

Attention User-defined data types are not fully supported yet. They are currently (as of Flink 1.11) only exposed as unregistered structured types in parameters and return types of functions.

A structured type is similar to an object in an object-oriented programming language. It contains zero, one or more attributes. Each attribute consists of a name and a type.

There are two kinds of structured types:

  • Types that are stored in a catalog and are identified by a catalog identifier (like cat.db.MyType). Those are equal to the SQL standard definition of structured types.

  • Anonymously defined, unregistered types (usually reflectively extracted) that are identified by an implementation class (like com.myorg.model.MyType). Those are useful when programmatically defining a table program. They enable reusing existing JVM classes without manually defining the schema of a data type again.

Registered Structured Types

Currently, registered structured types are not supported. Thus, they cannot be stored in a catalog or referenced in a CREATE TABLE DDL.

Unregistered Structured Types

Unregistered structured types can be created from regular POJOs (Plain Old Java Objects) using automatic reflective extraction.

The implementation class of a structured type must meet the following requirements:

  • The class must be globally accessible which means it must be declared public, static, and not abstract.
  • The class must offer a default constructor with zero arguments or a full constructor that assigns all fields.
  • All fields of the class must be readable by either public declaration or a getter that follows common coding style such as getField(), isField(), field().
  • All fields of the class must be writable by either public declaration, fully assigning constructor, or a setter that follows common coding style such as setField(...), field(...).
  • All fields must be mapped to a data type either implicitly via reflective extraction or explicitly using the @DataTypeHint annotations.
  • Fields that are declared static or transient are ignored.

The reflective extraction supports arbitrary nesting of fields as long as a field type does not (transitively) refer to itself.

The declared field class (e.g. public int age;) must be contained in the list of supported JVM bridging classes defined for every data type in this document (e.g. java.lang.Integer or int for INT).

For some classes an annotation is required in order to map the class to a data type (e.g. @DataTypeHint("DECIMAL(10, 2)") to assign a fixed precision and scale for java.math.BigDecimal).

Python

Declaration

Java

  1. class User {
  2. // extract fields automatically
  3. public int age;
  4. public String name;
  5. // enrich the extraction with precision information
  6. public @DataTypeHint("DECIMAL(10, 2)") BigDecimal totalBalance;
  7. // enrich the extraction with forcing using RAW types
  8. public @DataTypeHint("RAW") Class<?> modelClass;
  9. }
  10. DataTypes.of(User.class);

Bridging to JVM Types

Java TypeInputOutputRemarks
classXXOriginating class or subclasses (for input) or
superclasses (for output). Default
org.apache.flink.types.RowXXRepresent the structured type as a row.
org.apache.flink.table.data.RowDataXXInternal data structure.

Scala

  1. case class User(
  2. // extract fields automatically
  3. age: Int,
  4. name: String,
  5. // enrich the extraction with precision information
  6. @DataTypeHint("DECIMAL(10, 2)") totalBalance: java.math.BigDecimal,
  7. // enrich the extraction with forcing using a RAW type
  8. @DataTypeHint("RAW") modelClass: Class[_]
  9. )
  10. DataTypes.of(classOf[User])

Bridging to JVM Types

Java TypeInputOutputRemarks
classXXOriginating class or subclasses (for input) or
superclasses (for output). Default
org.apache.flink.types.RowXXRepresent the structured type as a row.
org.apache.flink.table.data.RowDataXXInternal data structure.

Python

  1. Not supported.

Other Data Types

BOOLEAN

Data type of a boolean with a (possibly) three-valued logic of TRUE, FALSE, and UNKNOWN.

Declaration

SQL

  1. BOOLEAN

Java/Scala

  1. DataTypes.BOOLEAN()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.BooleanXXDefault
booleanX(X)Output only if type is not nullable.

Python

  1. DataTypes.BOOLEAN()

RAW

Data type of an arbitrary serialized type. This type is a black box within the table ecosystem and is only deserialized at the edges.

The raw type is an extension to the SQL standard.

Declaration

SQL

  1. RAW('class', 'snapshot')

Java/Scala

  1. DataTypes.RAW(class, serializer)
  2. DataTypes.RAW(class)

Bridging to JVM Types

Java TypeInputOutputRemarks
classXXOriginating class or subclasses (for input) or
superclasses (for output). Default
byte[]X
org.apache.flink.table.data.RawValueDataXXInternal data structure.

Python

  1. Not supported.

SQL/Java/Scala

The type can be declared using RAW('class', 'snapshot') where class is the originating class and snapshot is the serialized TypeSerializerSnapshot in Base64 encoding. Usually, the type string is not declared directly but is generated while persisting the type.

In the API, the RAW type can be declared either by directly supplying a Class + TypeSerializer or by passing Class and letting the framework extract Class + TypeSerializer from there.

Python

NULL

Data type for representing untyped NULL values.

The null type is an extension to the SQL standard. A null type has no other value except NULL, thus, it can be cast to any nullable type similar to JVM semantics.

This type helps in representing unknown types in API calls that use a NULL literal as well as bridging to formats such as JSON or Avro that define such a type as well.

This type is not very useful in practice and is just mentioned here for completeness.

Declaration

SQL

  1. NULL

Java/Scala

  1. DataTypes.NULL()

Bridging to JVM Types

Java TypeInputOutputRemarks
java.lang.ObjectXXDefault
any class(X)Any non-primitive type.

Python

  1. Not supported.

Casting

Flink Table API and SQL can perform casting between a defined input type and target type. While some casting operations can always succeed regardless of the input value, others can fail at runtime (i.e. where there is no way to create a value for the target type). For example, it is always possible to convert INT to STRING, but you cannot always convert a STRING to INT.

During the planning stage, the query validator rejects queries for invalid type pairs with a ValidationException, e.g. when trying to cast a TIMESTAMP to an INTERVAL. Valid type pairs that can fail at runtime will be accepted by the query validator, but requires the user to correctly handle failures.

In Flink Table API and SQL, casting can be performed by using one of the two following built-in functions:

  • CAST: The regular cast function defined by the SQL standard. It can fail the job if the cast operation is fallible and the provided input is not valid. The type inference will preserve the nullability of the input type.
  • TRY_CAST: An extension to the regular cast function which returns NULL in case the cast operation fails. Its return type is always nullable.

For example:

  1. CAST('42' AS INT) --- returns 42 of type INT NOT NULL
  2. CAST(NULL AS VARCHAR) --- returns NULL of type VARCHAR
  3. CAST('non-number' AS INT) --- throws an exception and fails the job
  4. TRY_CAST('42' AS INT) --- returns 42 of type INT
  5. TRY_CAST(NULL AS VARCHAR) --- returns NULL of type VARCHAR
  6. TRY_CAST('non-number' AS INT) --- returns NULL of type INT
  7. COALESCE(TRY_CAST('non-number' AS INT), 0) --- returns 0 of type INT NOT NULL

The matrix below describes the supported cast pairs, where “Y” means supported, “!” means fallible, “N” means unsupported:

Input\TargetCHAR¹/
VARCHAR¹/
STRING
BINARY¹/
VARBINARY¹/
BYTES
BOOLEANDECIMALTINYINTSMALLINTINTEGERBIGINTFLOATDOUBLEDATETIMETIMESTAMPTIMESTAMP_LTZINTERVALARRAYMULTISETMAPROWSTRUCTUREDRAW
CHAR/
VARCHAR/
STRING
Y!!!!!!!!!!!!!NNNNNNN
BINARY/
VARBINARY/
BYTES
YYNNNNNNNNNNNNNNNNNNN
BOOLEANYNYYYYYYYYNNNNNNNNNNN
DECIMALYNNYYYYYYYNNNNNNNNNNN
TINYINTYNYYYYYYYYNNNNNNNNN
SMALLINTYNYYYYYYYYNNNNNNNNN
INTEGERYNYYYYYYYYNNY⁵NNNNNN
BIGINTYNYYYYYYYYNNY⁶NNNNNN
FLOATYNNYYYYYYYNNNNNNNNNNN
DOUBLEYNNYYYYYYYNNNNNNNNNNN
DATEYNNNNNNNNNYNYYNNNNNNN
TIMEYNNNNNNNNNNYYYNNNNNNN
TIMESTAMPYNNNNNNNNNYYYYNNNNNNN
TIMESTAMP_LTZYNNNNNNNNNYYYYNNNNNNN
INTERVALYNNNNNY⁵Y⁶NNNNNNYNNNNNN
ARRAYYNNNNNNNNNNNNNNNNNNN
MULTISETYNNNNNNNNNNNNNNNNNNN
MAPYNNNNNNNNNNNNNNNNNNN
ROWYNNNNNNNNNNNNNNNNNNN
STRUCTUREDYNNNNNNNNNNNNNNNNNNN
RAWY!NNNNNNNNNNNNNNNNNNY⁴

Notes:

  1. All the casting to constant length or variable length will also trim and pad accordingly to the type definition.
  2. TO_TIMESTAMP and TO_TIMESTAMP_LTZ must be used instead of CAST/TRY_CAST.
  3. Supported iff the children type pairs are supported. Fallible iff the children type pairs are fallible.
  4. Supported iff the RAW class and serializer are equals.
  5. Supported iff INTERVAL is a MONTH TO YEAR range.
  6. Supported iff INTERVAL is a DAY TO TIME range.

Also note that a cast of a NULL value will always return NULL, regardless of whether the function used is CAST or TRY_CAST.

Legacy casting

Pre Flink 1.15 casting behaviour can be enabled by setting table.exec.legacy-cast-behaviour to enabled. In Flink 1.15 this flag is disabled by default.

In particular, this will:

  • Disable trimming/padding for casting to CHAR/VARCHAR/BINARY/VARBINARY
  • CAST never fails but returns NULL, behaving as TRY_CAST but without inferring the correct type
  • Formatting of some casting to CHAR/VARCHAR/STRING produces slightly different results.

We discourage the use of this flag and we strongly suggest for new projects to keep this flag disabled and use the new casting behaviour. This flag will be removed in the next Flink versions.

Data Type Extraction

Java/Scala

At many locations in the API, Flink tries to automatically extract data type from class information using reflection to avoid repetitive manual schema work. However, extracting a data type reflectively is not always successful because logical information might be missing. Therefore, it might be necessary to add additional information close to a class or field declaration for supporting the extraction logic.

The following table lists classes that can be implicitly mapped to a data type without requiring further information.

If you intend to implement classes in Scala, it is recommended to use boxed types (e.g. java.lang.Integer) instead of Scala’s primitives. Scala’s primitives (e.g. Int or Double) are compiled to JVM primitives (e.g. int/double) and result in NOT NULL semantics as shown in the table below. Furthermore, Scala primitives that are used in generics (e.g. java.util.Map[Int, Double]) are erased during compilation and lead to class information similar to java.util.Map[java.lang.Object, java.lang.Object].

ClassData Type
java.lang.StringSTRING
java.lang.BooleanBOOLEAN
booleanBOOLEAN NOT NULL
java.lang.ByteTINYINT
byteTINYINT NOT NULL
java.lang.ShortSMALLINT
shortSMALLINT NOT NULL
java.lang.IntegerINT
intINT NOT NULL
java.lang.LongBIGINT
longBIGINT NOT NULL
java.lang.FloatFLOAT
floatFLOAT NOT NULL
java.lang.DoubleDOUBLE
doubleDOUBLE NOT NULL
java.sql.DateDATE
java.time.LocalDateDATE
java.sql.TimeTIME(0)
java.time.LocalTimeTIME(9)
java.sql.TimestampTIMESTAMP(9)
java.time.LocalDateTimeTIMESTAMP(9)
java.time.OffsetDateTimeTIMESTAMP(9) WITH TIME ZONE
java.time.InstantTIMESTAMP_LTZ(9)
java.time.DurationINTERVAL SECOND(9)
java.time.PeriodINTERVAL YEAR(4) TO MONTH
byte[]BYTES
T[]ARRAY<T>
java.util.Map<K, V>MAP<K, V>
structured type Tanonymous structured type T

Other JVM bridging classes mentioned in this document require a @DataTypeHint annotation.

Data type hints can parameterize or replace the default extraction logic of individual function parameters and return types, structured classes, or fields of structured classes. An implementer can choose to what extent the default extraction logic should be modified by declaring a @DataTypeHint annotation.

The @DataTypeHint annotation provides a set of optional hint parameters. Some of those parameters are shown in the following example. More information can be found in the documentation of the annotation class.

Python

Java

  1. import org.apache.flink.table.annotation.DataTypeHint;
  2. class User {
  3. // defines an INT data type with a default conversion class `java.lang.Integer`
  4. public @DataTypeHint("INT") Object o;
  5. // defines a TIMESTAMP data type of millisecond precision with an explicit conversion class
  6. public @DataTypeHint(value = "TIMESTAMP(3)", bridgedTo = java.sql.Timestamp.class) Object o;
  7. // enrich the extraction with forcing using a RAW type
  8. public @DataTypeHint("RAW") Class<?> modelClass;
  9. // defines that all occurrences of java.math.BigDecimal (also in nested fields) will be
  10. // extracted as DECIMAL(12, 2)
  11. public @DataTypeHint(defaultDecimalPrecision = 12, defaultDecimalScale = 2) AccountStatement stmt;
  12. // defines that whenever a type cannot be mapped to a data type, instead of throwing
  13. // an exception, always treat it as a RAW type
  14. public @DataTypeHint(allowRawGlobally = HintFlag.TRUE) ComplexModel model;
  15. }

Scala

  1. import org.apache.flink.table.annotation.DataTypeHint
  2. class User {
  3. // defines an INT data type with a default conversion class `java.lang.Integer`
  4. @DataTypeHint("INT")
  5. var o: AnyRef
  6. // defines a TIMESTAMP data type of millisecond precision with an explicit conversion class
  7. @DataTypeHint(value = "TIMESTAMP(3)", bridgedTo = java.sql.Timestamp.class)
  8. var o: AnyRef
  9. // enrich the extraction with forcing using a RAW type
  10. @DataTypeHint("RAW")
  11. var modelClass: Class[_]
  12. // defines that all occurrences of java.math.BigDecimal (also in nested fields) will be
  13. // extracted as DECIMAL(12, 2)
  14. @DataTypeHint(defaultDecimalPrecision = 12, defaultDecimalScale = 2)
  15. var stmt: AccountStatement
  16. // defines that whenever a type cannot be mapped to a data type, instead of throwing
  17. // an exception, always treat it as a RAW type
  18. @DataTypeHint(allowRawGlobally = HintFlag.TRUE)
  19. var model: ComplexModel
  20. }

Python

  1. Not supported.