SQL data types

info

Apache Druid supports two query languages: Druid SQL and native queries. This document describes the SQL language.

Druid associates each column with a specific data type. This topic describes supported data types in Druid SQL.

Standard types

Druid natively supports the following basic column types:

• LONG: 64-bit signed int
• FLOAT: 32-bit float
• DOUBLE: 64-bit float
• STRING: UTF-8 encoded strings and string arrays
• COMPLEX: non-standard data types, such as nested JSON, hyperUnique and approxHistogram, and DataSketches
• ARRAY: arrays composed of any of these types

Druid treats timestamps (including the __time column) as LONG, with the value being the number of milliseconds since 1970-01-01 00:00:00 UTC, not counting leap seconds. Therefore, timestamps in Druid do not carry any timezone information. They only carry information about the exact moment in time they represent. See Time functions for more information about timestamp handling.

The following table describes how Druid maps SQL types onto native types when running queries:

*The default value is NULL for all types, except in the deprecated legacy mode (druid.generic.useDefaultValueForNull = true) which initialize a default value.

For casts between two SQL types, the behavior depends on the runtime type:

Casts between two SQL types with the same Druid runtime type have no effect other than the exceptions noted in the table.

Casts between two SQL types that have different Druid runtime types generate a runtime cast in Druid.

If a value cannot be cast to the target type, as in CAST('foo' AS BIGINT), Druid a substitutes NULL. When druid.generic.useDefaultValueForNull = true (deprecated legacy mode), Druid instead substitutes a default value, including when NULL values cast to non-nullable types. For example, if druid.generic.useDefaultValueForNull = true, a null VARCHAR cast to BIGINT is converted to a zero.

Arrays

Druid supports ARRAY types, which behave as standard SQL arrays, where results are grouped by matching entire arrays. The UNNEST operator can be used to perform operations on individual array elements, translating each element into a separate row.

ARRAY typed columns can be stored in segments with JSON-based ingestion using the ‘auto’ typed dimension schema shared with schema auto-discovery to detect and ingest arrays as ARRAY typed columns. For SQL based ingestion, the query context parameter arrayIngestMode must be specified as "array" to ingest ARRAY types. In Druid 28, the default mode for this parameter is "mvd" for backwards compatibility, which instead can only handle ARRAY<STRING> which it stores in multi-value string columns.

You can convert multi-value dimensions to standard SQL arrays explicitly with MV_TO_ARRAY or implicitly using array functions. You can also use the array functions to construct arrays from multiple columns.

Druid serializes ARRAY results as a JSON string of the array by default, which can be controlled by the context parameter sqlStringifyArrays. When set to false and using JSON result formats, the arrays will instead be returned as regular JSON arrays instead of in stringified form.

Multi-value strings

Druid’s native type system allows strings to have multiple values. These multi-value string dimensions are reported in SQL as type VARCHAR and can be syntactically used like any other VARCHAR. Regular string functions that refer to multi-value string dimensions are applied to all values for each row individually.

You can treat multi-value string dimensions as arrays using special multi-value string functions, which perform powerful array-aware operations, but retain their VARCHAR type and behavior.

Grouping by multi-value dimensions observes the native Druid multi-value aggregation behavior, which is similar to an implicit SQL UNNEST. See Grouping for more information.

info

Because the SQL planner treats multi-value dimensions as VARCHAR, there are some inconsistencies between how they are handled in Druid SQL and in native queries. For instance, expressions involving multi-value dimensions may be incorrectly optimized by the Druid SQL planner. For example, multi_val_dim = 'a' AND multi_val_dim = 'b' is optimized to false, even though it is possible for a single row to have both 'a' and 'b' as values for multi_val_dim.

The SQL behavior of multi-value dimensions may change in a future release to more closely align with their behavior in native queries, but the multi-value string functions should be able to provide nearly all possible native functionality.

Multi-value strings behavior

The behavior of Druid multi-value string dimensions varies depending on the context of their usage.

When used with standard VARCHAR functions which expect a single input value per row, such as CONCAT, Druid will map the function across all values in the row. If the row is null or empty, the function receives NULL as its input.

When used with the explicit multi-value string functions, Druid processes the row values as if they were ARRAY typed. Any operations which produce null and empty rows are distinguished as separate values (unlike implicit mapping behavior). These multi-value string functions, typically denoted with an MV_ prefix, retain their VARCHAR type after the computation is complete. Note that Druid multi-value columns do not distinguish between empty and null rows. An empty row will never appear natively as input to a multi-valued function, but any multi-value function which manipulates the array form of the value may produce an empty array, which is handled separately while processing.

info

Do not mix the usage of multi-value functions and normal scalar functions within the same expression, as the planner will be unable to determine how to properly process the value given its ambiguous usage. A multi-value string must be treated consistently within an expression.

When converted to ARRAY or used with array functions, multi-value strings behave as standard SQL arrays and can no longer be manipulated with non-array functions.

By default Druid serializes multi-value VARCHAR results as a JSON string of the array, if grouping was not applied on the value. If the value was grouped, due to the implicit UNNEST behavior, all results will always be standard single value VARCHAR. ARRAY typed results serialization is controlled with the context parameter sqlStringifyArrays. When set to false and using JSON result formats, the arrays will instead be returned as regular JSON arrays instead of in stringified form.

NULL values

The druid.generic.useDefaultValueForNull runtime property controls Druid’s NULL handling mode. For the most SQL compliant behavior, set this to false (the default).

When druid.generic.useDefaultValueForNull = false (the default), NULLs are treated more closely to the SQL standard. In this mode, numeric NULL is permitted, and NULLs and empty strings are no longer treated as interchangeable. This property affects both storage and querying, and must be set on all Druid service types to be available at both ingestion time and query time. There is some overhead associated with the ability to handle NULLs; see the segment internals documentation for more details.

When druid.generic.useDefaultValueForNull = true (deprecated legacy mode), Druid treats NULLs and empty strings interchangeably, rather than according to the SQL standard. In this mode Druid SQL only has partial support for NULLs. For example, the expressions col IS NULL and col = '' are equivalent, and both evaluate to true if col contains an empty string. Similarly, the expression COALESCE(col1, col2) returns col2 if col1 is an empty string. While the COUNT(*) aggregator counts all rows, the COUNT(expr) aggregator counts the number of rows where expr is neither null nor the empty string. Numeric columns in this mode are not nullable; any null or missing values are treated as zeroes. This was the default prior to Druid 28.0.0, but will be removed in a future release so that Druid always behaves in an SQL compatible manner.

Boolean logic

By default, Druid uses SQL three-valued logic for filter processing and boolean expression evaluation. This behavior relies on three settings:

• druid.generic.useDefaultValueForNull must be set to false (default), a runtime property which allows NULL values to exist in numeric columns and expressions, and string typed columns to distinguish between NULL and the empty string
• druid.expressions.useStrictBooleans must be set to true (default), a runtime property controls Druid’s boolean logic mode for expressions, as well as coercing all expression boolean values to be represented with a 1 for true and 0 for false
• druid.generic.useThreeValueLogicForNativeFilters must be set to true (default), a runtime property which decouples three-value logic handling from druid.generic.useDefaultValueForNull and druid.expressions.useStrictBooleans for backwards compatibility with older versions of Druid that did not fully support SQL compatible null value logic handling

If any of these settings is configured with a non-default value, Druid will use two-valued logic for non-expression based filters. Expression based filters are controlled independently with druid.expressions.useStrictBooleans, which if set to false Druid will use two-valued logic for expressions.

These configurations have been deprecated and will be removed in a future release so that Druid always has SQL compliant behavior.

Nested columns

Druid supports storing nested data structures in segments using the native COMPLEX<json> type. See Nested columns for more information.

You can interact with nested data using JSON functions, which can extract nested values, parse from string, serialize to string, and create new COMPLEX<json> structures.

COMPLEX types have limited functionality outside the specialized functions that use them, so their behavior is undefined when:

• Grouping on complex values.
• Filtering directly on complex values.
• Used as inputs to aggregators without specialized handling for a specific complex type.

In many cases, functions are provided to translate COMPLEX value types to STRING, which serves as a workaround solution until COMPLEX type functionality can be improved.