Dialect parity between GoogleSQL and PostgreSQL

This page describes the dialect differences between GoogleSQL and PostgreSQL and offers recommendations for using PostgreSQL approaches for specific GoogleSQL features.

GoogleSQL dialect feature differences

GoogleSQL feature PostgreSQL dialect recommendation
Sample datasets No recommendation available.
BigQuery external datasets Use Spanner federated queries.
ENUM Use TEXT columns with checked constraints instead. Unlike ENUMS, the sort order of a TEXT column can't be user-defined. The following example restricts the column to only support the 'C', 'B', and 'A' values. 
CREATE TABLE singers (
 singer_id BIGINT PRIMARY KEY,
 type TEXT NOT NULL CHECK (type IN ('C', 'B', 'A'))
);
Graph No recommendation available.
HAVING MAX or HAVING MIN Use a JOIN or a subquery to filter for the MAX or MIN value for the aggregation. The following example requires filtering MAX or MIN in a subquery. 
WITH amount_per_year AS (
 SELECT 1000 AS amount, 2025 AS year
 UNION ALL
 SELECT 10000, 2024
 UNION ALL
 SELECT 500, 2023
 UNION ALL
 SELECT 1500, 2025
 UNION ALL
 SELECT 20000, 2024
)

SELECT SUM(amount) AS max_year_amount_sum
FROM amount_per_year
WHERE year = (SELECT MAX(year) FROM amount_per_year);
Informational foreign keys No recommendation available.
JSON data type Use the JSONB data type.
SELECT to_json(table) FROM table We recommend explicitly mapping each column with the jsonb_build_object function: 
WITH singers AS (
  SELECT 1::int8 AS id, 'Singer First Name'::text AS first_name
)

SELECT jsonb_build_object('id', id, 'first_name', first_name)
FROM singers;
ORDER BY … COLLATE … No recommendation available.
NUMERIC column as a primary key, secondary index, or foreign key We recommend using an index over a TEXT generated column, as shown in the following example: 
CREATE TABLE singers(
 id numeric NOT NULL,
 pk text GENERATED ALWAYS AS (id::text) STORED,
 PRIMARY KEY(pk)
);
Protocol buffer data type You can store serialized protocol buffers as the PostgreSQL BYTEA data type.
PRIMARY KEY DESC No recommendation available.
SELECT AS VALUE
SELECT * EXCEPT We recommend that you spell out all columns in the SELECT statement.
SELECT * REPLACE We recommend that you spell out all columns in the SELECT statement.
The following columns in the SPANNER_SYS statistics tables: No recommendation available.
TABLESAMPLE We recommend that you apply a custom function F, which converts a row to TEXT or BYTEA. You can then use spanner.farm_fingerprint to sample your data.

In the following example, we use CONCAT as our function F: 
-- Given the following schema

CREATE TABLE singers (
 singer_id BIGINT PRIMARY KEY,
 first_name VARCHAR(1024),
 last_name VARCHAR(1024),
 singer_info BYTEA
);

-- Create a hash for each row (using all columns)
WITH hashed_rows AS (
  SELECT
    *,
    ABS(MOD(spanner.farm_fingerprint(
      CONCAT(
        singer_id::text,
        first_name,
        last_name,
        singer_info::text
      )
    ), 100)) AS hash_value
  FROM singers
)

-- Sample data

SELECT *
FROM hashed_rows
WHERE hash_value < 10 -- sample roughly 10%
LIMIT 10; /* Optional: LIMIT to a max of 10 rows
             to be returned */
VALUE IN UNNEST(ARRAY(...)) Use the equality operator with the ANY function, as shown in the following example: 
SELECT value = any(array[...])

GoogleSQL dialect function differences

GoogleSQL function PostgreSQL dialect recommendation
ACOSH Use the formula of the function explicitly, as shown in the following example: 
SELECT LN(x + SQRT(x*x - 1));
APPROX_COSINE_DISTANCE No recommendation available.
APPROX_DOT_PRODUCT
APPROX_EUCLIDEAN_DISTANCE
ANY_VALUE Workaround available outside of aggregation and GROUP BY. Use a subquery with the ORDER BY or LIMIT clauses, as shown in the following example: 
SELECT * FROM
(
  (expression)
  UNION ALL SELECT NULL, … -- as many columns as you have
) AS rows
ORDER BY 1 NULLS LAST
LIMIT 1;
ARRAY_CONCAT_AGG You can use ARRAY_AGG and UNNEST as shown in the following example: 
WITH albums AS
(
  SELECT ARRAY['Song A', NULL, 'Song B'] AS songs
  UNION ALL
  SELECT NULL
  UNION ALL
  SELECT ARRAY[]::TEXT[]
)
SELECT ARRAY_AGG(song) FROM albums, UNNEST(songs) song;
ARRAY_FIRST Use the array subscript operator, as shown in the following example: 
SELECT array_expression[1];
 Note that this will return NULL for empty arrays.
ARRAY_INCLUDES Use the equality operator with the ANY function, as shown in the following example: 
SELECT search_value = ANY(array_to_search);
ARRAY_INCLUDES_ALL Use the array contains operator, as shown in the following example: 
SELECT array_to_search @> search_values;
ARRAY_INCLUDES_ANY Use the array overlap operator, as shown in the following example: 
SELECT array_to_search && search_values;
ARRAY_IS_DISTINCT Use a subquery to count distinct values and compare them to the original array length, as shown in the following example: 
SELECT ARRAY_LENGTH(value, 1) = (
SELECT COUNT(DISTINCT e)
FROM UNNEST(value) AS e);
ARRAY_LAST Use the array subscript operator, as shown in the following example 
SELECT (value)[ARRAY_LENGTH(value, 1)];
This returns NULL for empty arrays.
ARRAY_MAX Use a subquery with UNNEST and the MAX function, as shown in the following example: 
SELECT MAX(e) FROM UNNEST(value) AS e;
ARRAY_MIN Use a subquery with UNNEST and the MIN function, as shown in the following example: 
SELECT MIN(e) FROM UNNEST(value) AS e;
ARRAY_REVERSE No recommendation available.
ASINH Use the formula of the function explicitly, as shown in the following example: 
SELECT LN(x + SQRT(x*x - 1));
ATANH Use the formula of the function explicitly, as shown in the following example: 
SELECT 0.5 * LN((1 + x) / (1 - x));
BIT_COUNT No recommendation available.
BIT_XOR
BYTE_LENGTH
CODE_POINTS_TO_BYTES
CODE_POINTS_TO_STRING
COSH Use the formula of the function explicitly, as shown in the following example: 
SELECT (EXP(x) + EXP(-x)) / 2;
ERROR No recommendation available.
FROM_BASE32
FROM_BASE64
FROM_HEX
GENERATE_ARRAY
GENERATE_DATE_ARRAY
NET.HOST Use a regular expression and the substring function, as shown in the following example: 
/* Use modified regular expression from
  https://tools.ietf.org/html/rfc3986#appendix-A. */

SELECT Substring('http://www.google.com/test' FROM
  '^(?:[^:/?#]+:)?(?://)?([^/?#]*)?[^?#]*(?:\\?[^#]*)?(?:#.*)?')
NET.IP_FROM_STRING No recommendation available.
NET.IP_NET_MASK
NET.IP_TO_STRING
NET.IP_TRUNC
NET.IPV4_FROM_INT64
NET.IPV4_TO_INT64
NET.PUBLIC_SUFFIX
NET.REG_DOMAIN
NET.SAFE_IP_FROM_STRING
NORMALIZE
NORMALIZE_AND_CASEFOLD
REGEXP_EXTRACT_ALL
SAFE.ADD We recommend that you protect against an overflow explicitly leveraging the NUMERIC data type. 
WITH numbers AS
(
  SELECT 1::int8 AS a, 9223372036854775807::int8 AS b
  UNION ALL
  SELECT 1, 2
)

SELECT
 CASE
   WHEN a::numeric + b::numeric > 9223372036854775807 THEN NULL
   WHEN a + b < -9223372036854775808 THEN NULL
   ELSE a + b
 END AS result
FROM numbers;
SAFE.CAST No recommendation available.
SAFE.CONVERT_BYTES_TO_STRING
SAFE.DIVIDE We recommend that you protect against an overflow explicitly leveraging the NUMERIC data type during a division operation. 
WITH numbers AS
(
  SELECT 1::int8 AS a, 9223372036854775807::int8 AS b
  UNION ALL
  SELECT 10, 2
)

SELECT
 CASE
   WHEN b = 0 THEN NULL
   WHEN a::numeric / b::numeric > 9223372036854775807 THEN NULL
   WHEN a::numeric / b::numeric < -9223372036854775808 THEN NULL
   ELSE a / b
 END AS result
FROM numbers;
SAFE.MULTIPLY We recommend that you protect against an overflow explicitly leveraging the NUMERIC data type during a multiplication operation. 
WITH numbers AS
(
  SELECT 1::int8 AS a, 9223372036854775807::int8 AS b
  UNION ALL
  SELECT 1, 2
)

SELECT
 CASE
   WHEN a::numeric * b::numeric > 9223372036854775807 THEN NULL
   WHEN a::numeric * b::numeric < -9223372036854775808 THEN NULL
   ELSE a * b
 END AS result
FROM numbers;
SAFE.NEGATE We recommend that you protect against an overflow explicitly leveraging the NUMERIC data type during a negation operation. 
WITH numbers AS
(
  SELECT 9223372036854775807 AS a
  UNION ALL
  SELECT -9223372036854775808
)

SELECT
 CASE
   WHEN a <= -9223372036854775808 THEN NULL
   WHEN a >= 9223372036854775809 THEN NULL
   ELSE -a
 END AS result
FROM numbers;
SAFE.SUBTRACT We recommend that you protect against an overflow explicitly leveraging the NUMERIC data type during a subtraction operation. 
WITH numbers AS
(
  SELECT 1::int8 AS a, 9223372036854775807::int8 AS b
  UNION ALL
  SELECT 1, 2
)

SELECT
 CASE
   WHEN a::numeric - b::numeric > 9223372036854775807 THEN NULL
   WHEN a::numeric - b::numeric < -9223372036854775808 THEN NULL
   ELSE a - b
 END AS result
FROM numbers;
SAFE.TO_JSON No recommendation available.
SINH Use the formula of the function explicitly, as shown in the following example: 
SELECT (EXP(x) - EXP(-x)) / 2;
SPLIT Use the regexp_split_to_array function, as shown in the following example: 
WITH letters AS
(
  SELECT '' as letter_group
  UNION ALL
  SELECT 'a' as letter_group
  UNION ALL
  SELECT 'b c d' as letter_group
)

SELECT regexp_split_to_array(letter_group, ' ') as example
FROM letters;
STDDEV Use the formula of the function explicitly (unbiased standard deviation), as shown in the following example: 
WITH numbers AS
(
  SELECT 1 AS x
  UNION ALL
  SELECT 2
  UNION ALL
  SELECT 3
),

mean AS
(
  SELECT AVG(x)::float8 AS mean
  FROM numbers
)

SELECT SQRT(SUM(POWER(numbers.x - mean.mean, 2)) / (COUNT(x) - 1))
  AS stddev
FROM numbers
CROSS JOIN mean
STDDEV_SAMP Use the formula of the function explicitly (unbiased standard deviation), as shown in the following example: 
WITH numbers AS
(
  SELECT 1 AS x
  UNION ALL
  SELECT 2
  UNION ALL
  SELECT 3
),

mean AS (
  SELECT AVG(x)::float8 AS mean
  FROM numbers
)

SELECT SQRT(SUM(POWER(numbers.x - mean.mean, 2)) / (COUNT(x) - 1))
  AS stddev
FROM numbers
CROSS JOIN mean
TANH Use the formula of the function explicitly. 
SELECT (EXP(x) - EXP(-x)) / (EXP(x) + EXP(-x));
TIMESTAMP_MICROS Use the to_timestamp function and truncate the microseconds part of the input (precision loss), as shown in the following example: 
SELECT to_timestamp(1230219000123456 / 1000000);
TIMESTAMP_MILLIS Use the to_timestamp function and truncate the milliseconds part of the input (precision loss), as shown in the following example: 
SELECT to_timestamp(1230219000123 / 1000);
TO_BASE32 No recommendation available.
TO_BASE64
TO_CODE_POINTS
TO_HEX
VAR_SAMP Use the formula of the function explicitly (unbiased variance), as shown in the following: 
-- Use formula directly (unbiased)

WITH numbers AS
(
  SELECT 1 AS x
  UNION ALL
  SELECT 2
  UNION ALL
  SELECT 3 ), mean AS
(
  SELECT Avg(x)::float8 AS mean
  FROM   numbers )
SELECT Sum(Power(numbers.x - mean.mean, 2)) / (Count(x) - 1)
  AS variance
FROM numbers
CROSS JOIN mean
VARIANCE Use the formula of the function explicitly (unbiased variance), as shown in the following example: 
-- Use formula directly (unbiased VARIANCE like VAR_SAMP)

WITH numbers AS
(
  SELECT 1 AS x
  UNION ALL
  SELECT 2
  UNION ALL
  SELECT 3
),

mean AS (
  SELECT AVG(x)::float8 AS mean
  FROM numbers
)

SELECT SUM(POWER(numbers.x - mean.mean, 2)) / (COUNT(x) - 1)
  AS variance
FROM numbers
CROSS JOIN mean

What's next