druid/migr-ansi-sql-null.md at 0e6c388b7ffac4af710bf7499b03cf4efa26bff8

13 KiB

Raw Blame History

id	title	sidebar_label
migr-ansi-sql-null	Migration guide: SQL compliant mode	SQL compliant mode

import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem';

In Apache Druid 28.0.0, the default null handling mode changed to be compliant with the ANSI SQL standard. This guide provides strategies for Druid operators who rely on legacy Druid null handling behavior in their applications to transition to SQL compliant mode. Legacy mode is planned for removal from Druid.

SQL compliant null handling

As of Druid 28.0.0, Druid writes segments in an ANSI SQL compatible null handling mode by default. This means that Druid stores null values distinctly from empty strings for string dimensions and distinctly from 0 for numeric dimensions.

This can impact your application behavior because the ANSI SQL standard defines any comparison to null to be unknown. According to this three-valued logic, x <> 'some value' only returns non-null values.

The default Druid configurations for 28.0.0 and later that enable ANSI SQL compatible null handling mode are the following:

druid.generic.useDefaultValueForNull=false
druid.expressions.useStrictBooleans=true
druid.generic.useThreeValueLogicForNativeFilters=true

Follow the Null handling tutorial to learn how the default null handling works in Druid.

Legacy null handling and two-valued logic

Prior to Druid 28.0.0, Druid defaulted to a legacy mode which stored default values instead of nulls. In legacy mode, Druid created segments with the following characteristics at ingestion time:

String columns couldn't distinguish an empty string, '', from null. Therefore, Druid treated them both as interchangeable values.
Numeric columns couldn't represent null valued rows. Therefore, Druid stored 0 instead of null.

The Druid configurations for the deprecated legacy mode were the following:

druid.generic.useDefaultValueForNull=true
druid.expressions.useStrictBooleans=false
druid.generic.useThreeValueLogicForNativeFilters=true

These configurations are deprecated and scheduled for removal. After the configurations are removed, Druid will ignore them if they exist in your configuration files and use the default SQL compliant mode.

Migrate to SQL compliant mode

If your business logic relies on the behavior of legacy mode, you have the following options to operate Druid in an ANSI SQL compatible null handling mode:

Modify incoming data to either avoid nulls or avoid empty strings to achieve the same query behavior as legacy mode. This means modifying your ingestion SQL queries and ingestion specs to handle nulls or empty strings. For example, replacing a null for a string column with an empty string or a 0 for a numeric column. However, it means that your existing queries should operate as if Druid were in legacy mode. If you do not care about preserving null values, this is a good option for you.
Preserve null values and update all of your SQL queries to be ANSI SQL compliant. This means you can preserve the incoming data with nulls intact. However, you must rewrite any affected client-side queries to be ANSI SQL compliant. If you have a requirement to preserve null values, choose this option.

Replace null values at ingestion time

If you don't need to preserve null values within Druid, you can use a transform at ingestion time to replace nulls with other values.

Consider the following input data:

{"time":"2024-01-01T00:00:00.000Z","string_example":"my_string","number_example":99}
{"time":"2024-01-02T00:00:00.000Z","string_example":"","number_example":0}
{"time":"2024-01-03T00:00:00.000Z","string_example":null,"number_example":null}

The following example illustrates how to use COALESCE and NVL at ingestion time to avoid null values in Druid:

REPLACE INTO "no_nulls_example" OVERWRITE ALL
WITH "ext" AS (
  SELECT *
  FROM TABLE(
    EXTERN(
      '{"type":"inline","data":"{\"time\":\"2024-01-01T00:00:00.000Z\",\"string_example\":\"my_string\",\"number_example\":99}\n{\"time\":\"2024-01-02T00:00:00.000Z\",\"string_example\":\"\",\"number_example\":0}\n{\"time\":\"2024-01-03T00:00:00.000Z\",\"string_example\":null,\"number_example\":null}"}',
      '{"type":"json"}'
    )
  ) EXTEND ("time" VARCHAR, "string_example" VARCHAR, "number_example" BIGINT)
)
SELECT
  TIME_PARSE("time") AS "__time",
  -- Replace any null string values with an empty string
  COALESCE("string_example",'') AS string_example,
  -- Replace any null numeric values with 0
  NVL("number_example",0) AS number_example
FROM "ext"
PARTITIONED BY MONTH

{
  "type": "index_parallel",
  "spec": {
    "ioConfig": {
      "type": "index_parallel",
      "inputSource": {
        "type": "inline",
        "data": "{\"time\":\"2024-01-01T00:00:00.000Z\",\"string_example\":\"my_string\",\"number_example\":99}\n{\"time\":\"2024-01-02T00:00:00.000Z\",\"string_example\":\"\",\"number_example\":0}\n{\"time\":\"2024-01-03T00:00:00.000Z\",\"string_example\":null,\"number_example\":null}"
      },
      "inputFormat": {
        "type": "json"
      }
    },
    "tuningConfig": {
      "type": "index_parallel",
      "partitionsSpec": {
        "type": "dynamic"
      }
    },
    "dataSchema": {
      "dataSource": "inline_data_native",
      "timestampSpec": {
        "column": "time",
        "format": "iso"
      },
      "dimensionsSpec": {
        "dimensions": [
          "string_example",
          {
            "type": "long",
            "name": "number_example"
          }
        ]
      },
      "granularitySpec": {
        "queryGranularity": "none",
        "rollup": false,
        "segmentGranularity": "MONTH"
      },
      "transformSpec": {
        "transforms": [
          {
            "type": "expression",
            "name": "string_example",
            "expression": "COALESCE(\"string_example\",'')"
          },
          {
            "type": "expression",
            "name": "number_example",
            "expression": "NVL(\"number_example\",0)"
          }
        ]
      }
    }
  }
}

Druid ingests the data with no null values as follows:

`__time`	`string_example`	`number_example`
`2024-01-01T00:00:00.000Z`	`my_string`	99
`2024-01-02T00:00:00.000Z`	`empty`	0
`2024-01-03T00:00:00.000Z`	`empty`	0

Coerce empty strings to null at ingestion time

In legacy mode, Druid recognized empty strings as nulls for equality comparison. If your queries rely on empty strings to represent nulls, you can coerce empty strings to null at ingestion time using NULLIF.

For example, consider the following sample input data:

{"time":"2024-01-01T00:00:00.000Z","string_example":"my_string"}
{"time":"2024-01-02T00:00:00.000Z","string_example":""}
{"time":"2024-01-03T00:00:00.000Z","string_example":null}

In legacy mode, Druid wrote an empty string for the third record. Therefore the following query returned 2:

SELECT count(*)
FROM "null_string"
WHERE "string_example" IS NULL

In SQL compliant mode, Druid differentiates between empty strings and nulls, so the same query would return 1. The following example shows how to coerce empty strings into null to accommodate IS NULL comparisons:

REPLACE INTO "null_string" OVERWRITE ALL
WITH "ext" AS (
  SELECT *
  FROM TABLE(
    EXTERN(
      '{"type":"inline","data":"{\"time\":\"2024-01-01T00:00:00.000Z\",\"string_example\":\"my_string\"}\n{\"time\":\"2024-01-02T00:00:00.000Z\",\"string_example\":\"\"}\n{\"time\":\"2024-01-03T00:00:00.000Z\",\"string_example\":null}"}',
      '{"type":"json"}'
    )
  ) EXTEND ("time" VARCHAR, "string_example" VARCHAR)
)
SELECT
  TIME_PARSE("time") AS "__time",
  NULLIF("string_example",'') AS "string_example"
FROM "ext"
PARTITIONED BY MONTH

{
  "type": "index_parallel",
  "spec": {
    "ioConfig": {
      "type": "index_parallel",
      "inputSource": {
        "type": "inline",
        "data": "{\"time\":\"2024-01-01T00:00:00.000Z\",\"string_example\":\"my_string\"}\n{\"time\":\"2024-01-02T00:00:00.000Z\",\"string_example\":\"\"}\n{\"time\":\"2024-01-03T00:00:00.000Z\",\"string_example\":null}"
      },
      "inputFormat": {
        "type": "json"
      }
    },
    "tuningConfig": {
      "type": "index_parallel",
      "partitionsSpec": {
        "type": "dynamic"
      }
    },
    "dataSchema": {
      "dataSource": "null_string",
      "timestampSpec": {
        "column": "time",
        "format": "iso"
      },
      "transformSpec": {
        "transforms": [
          {
            "type": "expression",
            "expression": "case_searched((\"string_example\" == ''),null,\"string_example\")",
            "name": "string_example"
          }
        ]
      },
      "dimensionsSpec": {
        "dimensions": [
          "string_example"
        ]
      },
      "granularitySpec": {
        "queryGranularity": "none",
        "rollup": false,
        "segmentGranularity": "month"
      }
    }
  }
}

Druid ingests the data with no empty strings as follows:

`__time`	`string_examle`
`2024-01-01T00:00:00.000Z`	`my_string`
`2024-01-02T00:00:00.000Z`	`null`
`2024-01-03T00:00:00.000Z`	`null`

Therefore SELECT count(*) FROM "null_string" WHERE "string_example" IS NULL returns 2.

Rewrite your queries to be SQL compliant

If you want to maintain null values in your data within Druid, you can use the following ANSI SQL compliant querying strategies to achieve the same results as legacy null handling:

Modify inequality queries to include null values. For example, x <> 'some value' becomes (x <> 'some value' OR x IS NULL).
Use COALESCE or NVL to replace nulls with a value. For example, x + 1 becomes NVL(numeric_value, 0)+1

Consider the following Druid datasource null_example:

`__time`	`string_examle`	`number_example`
`2024-01-01T00:00:00.000Z`	`my_string`	99
`2024-01-02T00:00:00.000Z`	`empty`	0
`2024-01-03T00:00:00.000Z`	`null`	null

Druid excludes null strings from equality comparisons. For example:

SELECT COUNT(*) AS count_example
FROM "null_example"
WHERE "string_example"<> 'my_string'

Druid returns 1 because null is considered unknown: neither equal nor unequal to the value.

To count null values in the result, you can use an OR operator:

SELECT COUNT(*) AS count_example
FROM "null_example"
WHERE ("string_example"<> 'my_string') OR "string_example" IS NULL

Druid returns 2. To achieve the same result, you can use IS DISTINCT FROM for null-safe comparison:

SELECT COUNT(*) as count_example
FROM "null_example"
WHERE "string_example" IS DISTINCT FROM 'my_string'

Similarly, arithmetic operators on null return null. For example:

SELECT "number_example" + 1 AS additon_example
FROM "null_example"

Druid returns the following because null + any value is null for the ANSI SQL standard:

`addition_example`
100
1
null

Use NVL to avoid nulls with arithmetic. For example:

SELECT NVL("number_example",0) + 1 AS additon_example
FROM "null_example"

Druid returns the following:

`addition_example`
100
1
1

Learn more

See the following topics for more information:

Null handling tutorial to learn how the default null handling works in Druid.
Null values for a description of Druid's behavior with null values.
Handling null values for details about how Druid stores null values.

13 KiB Raw Blame History