GoogleSQL data definition language

GoogleSQL is the new name for Google Standard SQL! New name, same great SQL dialect.

Use GoogleSQL's data definition language (DDL) to:

• Create a database.
• Create, alter, or drop tables in a database.
• Add, alter, or drop columns in a table.
• Create or drop indexes in a database.
• Create, replace, or drop views in a database.
• Create, alter, or drop change streams in a database.
• Create or drop database roles.
• Grant privileges to database roles.
• Grant database roles to other database roles.
• Create, alter, or drop ML models in a database.

Notation

• Square brackets "[ ]" indicate optional clauses.
• Parentheses "( )" indicate literal parentheses.
• The vertical bar "|" indicates a logical OR.
• Curly braces "{ }" enclose a set of options.
• A comma followed by an ellipsis indicates that the preceding item can repeat in a comma-separated list. item [, ...] indicates one or more items, and [item, ...] indicates zero or more items.
• A comma "," indicates the literal comma.
• Angle brackets "<>" indicate literal angle brackets.
• An mdash "—" indicates a range of values between the items on either side of it.
• The plus sign "+" indicates that the preceding item can repeat.

Reserved keywords

Some words have special meaning in the GoogleSQL language and are reserved in its DDL. To use a reserved keyword as an identifier in your schema, enclose it in backticks (`). For the full list of reserved keywords in GoogleSQL, see GoogleSQL lexical structure and syntax.

For example:

CREATE TABLE MyTable (
  RowId INT64 NOT NULL,
  `Order` INT64
) PRIMARY KEY (RowId);

Naming conventions

The following rules apply to database IDs.

• Must start with a lowercase letter.
• Can contain lowercase letters, numbers, underscores, and hyphens, but not uppercase letters.
• Cannot end with an underscore or hyphen.
• Must be enclosed in backticks (`) if it's a reserved word or contains a hyphen.
• Can be between 2-30 characters long.
• Cannot be changed after you create it.

The following rules apply to table, column, index, view, role, constraint, and sequence names.

• Must be at least one character long.

• Can contain a maximum of 128 characters.

• Must start with an uppercase or lowercase letter.

• Can contain uppercase and lowercase letters, numbers, and underscores, but not hyphens.

• No two Spanner objects can be created with the same name, including names that differ only in capitalization. For example, the second statement in the following snippet fails because the table names differ only by case.

  CREATE TABLE MyTable (col1 INT64) PRIMARY KEY (col1);
  CREATE TABLE MYTABLE (col1 INT64) PRIMARY KEY (col1);
  

• When referring to other schema objects in a DDL statement (for example, a column name for a primary key, or table and column names in an index), make sure to use the original case for the name of each entity. As an example, consider the table Singers created with the following statement.

  CREATE TABLE Singers (
    SingerId   INT64 NOT NULL,
    FirstName  STRING(1024),
    LastName   STRING(1024),
    SingerInfo BYTES(MAX),
    BirthDate  DATE
  ) PRIMARY KEY(SingerId);
  

  The following command fails with the message Table not found: singers because it uses a different case for the Singers table.

  CREATE INDEX SingersByFirstLastName ON singers(FirstName, LastName)
  

• Schema object names are case insensitive in SQL queries. As an example, consider the table MyTable2 created with the following statement.

  CREATE TABLE MyTable2 (col1 INT64) PRIMARY KEY (col1);
  

  The following queries all succeed because schema object names are case-insensitive for queries.

  SELECT col1 FROM MyTable2 LIMIT 1;
  SELECT COL1 FROM MYTABLE2 LIMIT 1;
  SELECT COL1 FROM mytable2 LIMIT 1;
  INSERT INTO MYTABLE2 (col1) VALUES(1);
  

Data types

The following are the data types used in GoogleSQL.

Scalars

The syntax for using a scalar type in DDL is:

{
  BOOL
  | INT64
  | FLOAT64
  | NUMERIC
  | STRING( length )
  | JSON
  | BYTES( length )
  | DATE
  | TIMESTAMP
}

length:
    { int64_value | MAX }

int64_value:
    { decimal_value | hex_value }

decimal_value:
    [-]0—9+

hex_value:
    [-]0x{0—9|a—f|A—F}+

An int64_value must correspond to an integer from -9,223,372,036,854,775,808 (-2⁶³) to 9,223,372,036,854,775,807 (2⁶³ − 1). It can be specified with decimal or hexadecimal notation. The hexadecimal form requires a 0x prefix, with a lowercase x.

STRING

STRING is a variable length Unicode character string. Its value must be a valid Unicode string. Length is required, and represents the maximum number of Unicode characters (not bytes) that can be stored in the field.

Notes:

• Writes to the column are rejected if the new value is not a valid Unicode string or exceeds the specified length.

• length can be an integer in the range [1, 2621440].

• For a field whose length is unpredictable or does not need to be constrained, you can set length to the convenience value MAX, which is equivalent to 2621440 for validation purposes.

  Only the actual length of the stored string impacts storage costs; specifying MAX does not use any additional storage capacity.

• GoogleSQL requires Unicode strings to be UTF-8 encoded on receipt at the server.

• Collation is done by Unicode character numerical value (technically by code point, which is subtly different due to combining characters). For ASCII strings, this is the traditional sort order.

• You can reduce the length of a column after the table has been created, but doing so requires Spanner to validate that the existing data is within the length constraint.

JSON

JSON is a variable length Unicode character string representing a JSON object. The string must be UTF-8 encoded on receipt at the server. The maximum length of the JSON value is 2621440 characters.

See Working with JSON and Data types for details.

BYTES

BYTES is a variable length binary string. Length is required, and represents the maximum number of bytes that can be stored in the field.

Notes:

• Writes to the column are rejected if the new value exceeds the specified length.

• length can be an integer in the range [1, 10485760] or the convenience value MAX, which is equivalent to 10485760 for validation purposes.

  Only the actual stored bytes impact storage costs; specifying MAX does not use any additional storage capacity.

• You can reduce the length of a column after the table has been created, but doing so requires Spanner to validate that the existing data is within the length constraint.

DATE

• A timezone-independent date.
• The range [0001-01-01, 9999-12-31] is the legal interval for dates. A write to a date column is rejected if the value is outside of that interval.
• See more information and the canonical format in Data Types.

TIMESTAMP

• A timestamp with nanosecond precision.
• Timezone-independent, over the range [0001-01-01 00:00:00 to 10000-01-01 00:00:00).
• See more information and the canonical format in Data Types.

Arrays

The syntax for using the ARRAY type in DDL is:

ARRAY< scalar_type >

GoogleSQL supports arrays of scalars. The primary purpose of arrays is to store a collection of values in a space efficient way. Arrays are not designed to provide access to individual elements; to read or write a single element, you must read or write the entire array.

If your application uses data structures like vectors or repeated fields, you can persist their state in a GoogleSQL array.

Here's an example of an alternate definition of Singers that uses multiple columns of ARRAY type:

CREATE TABLE Singers (
  SingerId INT64,
  FeaturedSingerIds ARRAY<INT64>,
  SongNames ARRAY<STRING(MAX)>
) PRIMARY KEY (SingerId) ...;

Notes:

• Arrays with subtype ARRAY (nested arrays) are not supported.
• Arrays, like scalar values, can never be larger than 10 MiB total.
• Arrays can't be used as key columns.

• In a CREATE TABLE statement, you can create columns of ARRAY type with a NOT NULL annotation.

  After you create the table, you cannot add a column of ARRAY type with a NOT NULL annotation, and you cannot add a NOT NULL annotation to an existing column of ARRAY type.

DATABASE statements

CREATE DATABASE

When creating a GoogleSQL database, you must provide a CREATE DATABASE statement, which defines the ID of the database:

CREATE DATABASE database_id

where database_id
    {a—z}[{a—z|0—9|_|-}+]{a—z|0—9}

Parameters

database_id

• The name of the database to create. See Naming Conventions.

ALTER DATABASE

Changes the definition of a database.

Syntax

ALTER DATABASE database_id
    action

where database_id is:
    {a—z}[{a—z|0—9|_|-}+]{a—z|0—9}

and action is:
    SET OPTIONS ( options_def [, ... ] )

and options_def is:
    { default_leader = { 'region' | null } |
      optimizer_version = { 1 ... 6 | null } |
      optimizer_statistics_package = { 'package_name' | null } |
      version_retention_period = { 'duration' | null } }

Description

ALTER DATABASE changes the definition of an existing database.

SET OPTIONS

• Use this clause to set an option at the database level of the schema hierarchy.

Parameters

database_id

• The name of the database whose attributes are to be altered. If the name is a reserved word or contains a hyphen, enclose it in backticks (`). For more information on database naming conventions, see Naming Conventions in this document.

options_def

• The optimizer_version = { 1 ... 6 | null } option allows you to specify the query optimizer version to use. Setting this option to null is equivalent to setting it to the default version. For more information, see Query Optimizer.

• The optimizer_statistics_package = { 'package_name' | null } option allows you to specify the query optimizer statistics package name to use. By default, this is the latest collected statistics package, but you can specify any available statistics package version. Setting this option to null is equivalent to setting it to the latest version. For more information, see Query statistics package versioning.

• The version_retention_period = { 'duration' | null } is the period for which Spanner retains all versions of data and schema for the database. The duration must be in the range [1h, 7d] and can be specified in days, hours, minutes, or seconds. For example, the values 1d, 24h, 1440m, and 86400s are equivalent. Setting the value to null resets the retention period to the default, which is 1 hour. This option can be used for point-in-time recovery. For more information, see Point-in-time Recovery.

• The default_leader = { 'region' | null } sets the leader region for your database. You can only use this parameter for databases that use a multi-region configuration. default_leader must be set to null, or one of the read-write replicas in your multi-region configuration. null resets the leader region to the default leader region for your database's multi-region configuration. For more information, see Configuring the default leader region.

TABLE statements

CREATE TABLE

Defines a new table.

Syntax

CREATE TABLE [ IF NOT EXISTS ] table_name ( [
   { column_name data_type [NOT NULL]
     [ { DEFAULT ( expression ) | AS ( expression ) STORED } ] [ options_def ]
     | table_constraint }
   [, ... ]
] ) PRIMARY KEY ( [column_name [ { ASC | DESC } ], ...] )
[, INTERLEAVE IN PARENT table_name [ ON DELETE { CASCADE | NO ACTION } ] ]
[, ROW DELETION POLICY ( OLDER_THAN ( timestamp_column, INTERVAL num_days DAY ) ) ]

where data_type is:
    { scalar_type | array_type }

and options_def is:
    { OPTIONS ( allow_commit_timestamp = { true | null } ) }

and table_constraint is:
    [ CONSTRAINT constraint_name ]
    { CHECK ( expression ) |
      FOREIGN KEY ( column_name [, ... ] ) REFERENCES  ref_table  ( ref_column [, ... ] )
        [ ON DELETE { CASCADE | NO ACTION } ]
    }

Description

CREATE TABLE defines a new table in the current database.

Parameters

IF NOT EXISTS

• If a table exists with the same name, the CREATE statement has no effect and no error is generated.

table_name

• The name of the table to be created. For naming guidance, see Naming conventions.

column_name

• The name of a column to be created. For naming guidance, see Naming conventions.

data_type

• The data type of the column, which can be a Scalar or an Array type.

timestamp_column

• The name of a column of type TIMESTAMP, that is also specified in the CREATE TABLE statement.

num_days

• The number of days after the date in the specified timestamp_column, after which the row is marked for deletion. Valid values are non-negative integers.

NOT NULL

• This optional column annotation specifies that the column is required for all mutations that insert a new row.

• You cannot add a NOT NULL column to an existing table. For most column types, you can work around this limitation:

  • For columns of ARRAY type, the only time you can use a NOT NULL annotation is when you create the table. After that, you cannot add a NOT NULL annotation to a column of ARRAY type.

  • For all other column types, you can add a nullable column; fill that column by writing values to all rows; and update your schema with a NOT NULL annotation on that column.

DEFAULT (expression)

• This clause sets a default value for the column.
• A column with a default value can be a key or non-key column.
• A column can't have a default value and also be a generated column.
• You can insert your own value into a column that has a default value, overriding the default value. You can also reset a non-key column to its default value by using UPDATE ... SETcolumn-name= DEFAULT.
• A generated column or a check constraint can depend on a column with a default value.
• A column with a default value can't be a commit timestamp column.
  • PENDING_COMMIT_TIMESTAMP() can't be used as a default value.
  • SET OPTIONS (allow_commit_timestamp = true) is disallowed.

• expression can be a literal or any valid SQL expression that is assignable to the column data type, with the following properties and restrictions:

  • The expression can be non-deterministic.
  • The expression can't reference other columns.
  • The expression can't contain subqueries, query parameters, aggregates, or analytic functions.

AS (expression) STORED

• This clause creates a column as a generated column, which is a column whose value is defined as a function of other columns in the same row.

• expression can be any valid SQL expression that's assignable to the column data type with the following restrictions.

  • The expression can only reference columns in the same table.

  • The expression can't contain subqueries.

  • The expression can't contain non-deterministic functions such as PENDING_COMMIT_TIMESTAMP(), CURRENT_DATE(), and CURRENT_TIMESTAMP().

  • You can't modify the expression of a generated column.

• The STORED attribute that follows the expression stores the result of the expression along with other columns of the table. Subsequent updates to any of the referenced columns cause Spanner to re-evaluate and store the expression.

• Spanner doesn't allow generated columns unless they use the STORED attribute.

• Direct writes to generated columns aren't allowed.

• Column option allow_commit_timestamp isn't allowed on generated columns or any columns that generated columns reference.

• You can't change the data type of a generated column, or of any columns the generated column references.

• You can't drop a column a generated column references.

• You can use a generated column as a primary key with the following additional restrictions:

  • The generated primary key can't reference other generated columns.

  • The generated primary key can reference, at most, one non-key column.

  • The generated primary key can't depend on a non-key column with a DEFAULT clause.

• The following rules apply when using generated key columns:

  • Read APIs: You must fully specify the key columns, including the generated key columns.
  • Mutation APIs: For INSERT, INSERT_OR_UPDATE, and REPLACE, Spanner doesn't allow you to specify generated key columns. For UPDATE, you can optionally specify generated key columns. For DELETE, you need to fully specify the key columns including the generated keys.
  • DML: You can't explicitly write to generated keys in INSERT or UPDATE statements.
  • Query: In general, we recommend that you use the generated key column as a filter in your query. Optionally, if the expression for the generated key column uses only one column as a reference, the query can apply an equality (=) or IN condition to the referenced column. For more information and an example, see Create a unique key derived from a value column.

For examples on how to work with generated columns, see Creating and managing generated columns.

PRIMARY KEY ( [column_name[ { ASC | DESC } ], ...]

• Every table must have a primary key, and that primary key can be composed of zero or more columns of that table.

• Adding the DESC annotation on a primary key column name changes the physical layout of data from ascending order (default) to descending order.

  For more details, see Schema and data model.

[, INTERLEAVE IN PARENTtable_name[ ON DELETE { CASCADE | NO ACTION } ] ]

• This clause defines a child-to-parent table relationship, which results in a physical interleaving of parent and child rows. The primary-key columns of a parent must positionally match, both in name and type, a prefix of the primary-key columns of any child. Adding rows to the child table fails if the corresponding parent row does not exist. The parent row can either exist in the database or be inserted before the insertion of the child rows in the same transaction.

• The optional ON DELETE clause defines the behavior of rows in ChildTable when a mutation attempts to delete the parent row. The supported options are:

  • CASCADE: the child rows are deleted.

  • NO ACTION: the child rows are not deleted. If deleting a parent would leave behind child rows, thus violating parent-child referential integrity, the write will fail.

  You can omit the ON DELETE clause, in which case the default of ON DELETE NO ACTION is used.

    For more details, see Schema and data model.

CONSTRAINTconstraint_name

• An optional name for a table constraint. If a name is not specified, Spanner generates a name for the constraint. Constraints names, including generated names, can be queried from the Spanner information schema.

CHECK (expression)

• A CHECK constraint lets you specify that the values of one or more columns must satisfy a boolean expression.

• expression can be any valid SQL expression that evaluates to a BOOL.

• The following restrictions apply to a check constraint expression term.

  • The expression can only reference columns in the same table.

  • The expression must reference at least one non-generated column, whether directly or through a generated column which references a non-generated column.

  • The expression can't reference columns that have set the allow_commit_timestamp option.

  • The expression can't contain subqueries.

  • The expression can't contain non-deterministic functions, such as CURRENT_DATE() and CURRENT_TIMESTAMP().

• For more information, see Creating and managing check constraints.

FOREIGN KEY (column_name[, ... ] ) REFERENCESref_table(ref_column[, ... ] [ ON DELETE { CASCADE | NO ACTION } ] )

• Use this clause to define a foreign key constraint. A foreign key is defined on the referencing table of the relationship, and it references the referenced table. The foreign key columns of the two tables are called the referencing and referenced columns, and their row values are the keys.

• A foreign key constraint requires that one or more columns of this table can contain only values that are in the referenced columns of the referenced table.

• When creating a foreign key, a unique constraint is automatically created on the referenced table, unless the entire primary key is referenced. If the unique constraint can't be satisfied, the entire schema change will fail.

• The number of referencing and referenced columns must be the same. Order is also significant. That is, the first referencing column refers to the first referenced column, the second to the second, and so on.

• The referencing and referenced columns must have matching types and they must support the equality operator ('='). The columns must also be indexable. Columns of type ARRAY are not allowed.

• When you create a foreign key with the ON DELETE CASCADE action, deleting a row in the referenced table atomically deletes all rows from the referencing table that references the deleted row in the same transaction.

• If you do not specify a foreign key action, the default action is NO ACTION.

• Foreign keys can't be created on columns with the allow_commit_timestamp=true option.

    For details, see Foreign keys.

OPTIONS ( allow_commit_timestamp = { true | null } )

• The allow_commit_timestamp option allows insert and update operations to request that Spanner write the commit timestamp of the transaction into the column. For details, see Commit timestamps in GoogleSQL-dialect databases.

[, ROW DELETION POLICY ( OLDER_THAN (timestamp_column, INTERVALnum_daysDAY ) ) ]

• Use this clause to set a row deletion policy for this table. See Time to live (TTL) for details.

ALTER TABLE

Changes the definition of a table.

Syntax

ALTER TABLE table_name
    action

where action is:
    ADD [ COLUMN ] [ IF NOT EXISTS] column_name data_type [ column_expression ] [ options_def ]
    DROP [ COLUMN ] column_name
    ADD table_constraint
    DROP CONSTRAINT constraint_name
    SET ON DELETE { CASCADE | NO ACTION }
    ALTER [ COLUMN ] column_name
      {
        data_type  [ NOT NULL ] [ DEFAULT ( expression ) ]
        | SET  ( options_def )
        | SET  DEFAULT ( expression )
        | DROP DEFAULT
      }
    ADD ROW DELETION POLICY ( OLDER_THAN ( timestamp_column, INTERVAL num_days DAY ))
    DROP ROW DELETION POLICY
    REPLACE ROW DELETION POLICY ( OLDER_THAN ( timestamp_column, INTERVAL num_days DAY ))

and data_type is:
    { scalar_type | array_type }

and column_expression is:
    [ NOT NULL ] [ { DEFAULT ( expression ) | AS ( expression ) STORED } ]

and options_def is:
    OPTIONS (allow_commit_timestamp = { true | null } )

and table_constraint is:
    [ CONSTRAINT constraint_name ]
    { CHECK ( expression ) |
      FOREIGN KEY ( column_name [, ... ] ) REFERENCES ref_table ( ref_column [, ... ] )
        [ ON DELETE { CASCADE | NO ACTION } ]
    }

Description

ALTER TABLE changes the definition of an existing table.

ADD COLUMN

• Adds a new column to the table, using the same syntax as CREATE TABLE.

• If you specify IF NOT EXISTS and a column of the same name already exists, the statement has no effect and no error is generated.

• You can specify NOT NULL in an ALTER TABLE...ADD COLUMN statement if you specify DEFAULT (expression) or AS (expression) STORED for the column.

• If you include DEFAULT (expression) or AS (expression) STORED, the expression is evaluated and the computed value is backfilled for existing rows. The backfill operation is asynchronous. This backfill operation happens only when an ADD COLUMN statement is issued. There's no backfill on ALTER COLUMN.

• The DEFAULT clause has restrictions. See the description of this clause in CREATE TABLE.

DROP COLUMN

• Drops a column from a table.

• You can't drop a column referenced by a generated column.

• Dropping a column referenced by a CHECK constraint is not allowed.

ADDtable_constraint

• Adds a new constraint to a table using the same syntax as CREATE TABLE.

• For foreign keys, the existing data is validated before the foreign key is added. If any existing constrained key does not have a corresponding referenced key, or the referenced key is not unique, the constraint is in violation, and the ALTER statement fails.

• Adding a foreign key action on an existing foreign key constraint isn't supported. You need to add a new foreign key constraint with action.

• If you do not specify a foreign key action, the default action is NO ACTION.

• For CHECK constraints, new data is validated immediately against the constraint. A long-running process is also started to validate the existing data against the constraint. If any existing data does not conform to the constraint, the check constraint is rolled back.

• The following restrictions apply to a check constraint expression term.

  • The expression can only reference columns in the same table.

  • The expression must reference at least one non-generated column, whether directly or through a generated column which references a non-generated column.

  • The expression can't reference columns that have set the allow_commit_timestamp option.

  • The expression can't contain subqueries.

  • The expression can't contain non-deterministic functions, such as CURRENT_DATE() and CURRENT_TIMESTAMP().

DROP CONSTRAINTconstraint_name

• Drops the specified constraint on a table, along with any associated index, if applicable.

SET ON DELETE { CASCADE | NO ACTION }

• This alteration can be applied only on child tables of parent-child, interleaved tables relationships. For more information, see Schema and data model.

• The ON DELETE CASCADE clause signifies that when a row from the parent table is deleted, its child rows in this table will automatically be deleted as well. Child rows are all rows that start with the same primary key. If a child table does not have this annotation, or the annotation is ON DELETE NO ACTION, then you must delete the child rows before you can delete the parent row.

ALTER COLUMN

• Changes the definition of an existing column on a table.

• data_type[ NOT NULL ] [ DEFAULT (expression) ]

  • This clause changes the data type of the column and optionally sets a default value for the column.

  • The DEFAULT clause has restrictions. See the description of this clause in CREATE TABLE.

  • Statements to set, change, or drop default value of an existing column do not affect existing rows.

  • If the column has data and is altered to have the NOT NULL constraint, the statement might fail if there is at least one existing row with a NULL value. This is true even when a NOT NULL DEFAULT (...) is specified, because there is no backfill operation for ALTER COLUMN.

• SET (options_def)

  • The OPTIONS (allow_commit_timestamp = { true | null } ) clause is the only allowed option. If true, a commit timestamp can be stored into the column.

    To learn about commit timestamps, see Commit timestamps in GoogleSQL-dialect databases.

• SET DEFAULT( expression )

  • Sets or changes a default value for the column. Only metadata is affected. Existing data is not changed.

  • This clause has restrictions. See the description of this clause in CREATE TABLE.

  • When you use this clause, the expression result must be assignable to the current column type. To change the column type and default value in a single statement, use:

    ALTER TABLEtable-nameALTER COLUMNcolumn-name data_typeDEFAULTexpression

• DROP DEFAULT

  • Drops the column default value. Only metadata is affected. Existing data is not changed.

ADD ROW DELETION POLICY ( OLDER_THAN (timestamp_column, INTERVALnum_daysDAY ) )

• Adds a row deletion policy to the table defining the amount of time after a specific date after which to delete a row. See Time to live. Only one row deletion policy can exist on a table at a time.

DROP ROW DELETION POLICY

• Drops the row deletion policy on a table.

REPLACE ROW DELETION POLICY ( OLDER_THAN (timestamp_column, INTERVALnum_daysDAY ) )

• Replaces the existing row deletion policy with a new policy.

Parameters

table_name

• The name of an existing table to alter.

column_name

• The name of a new or existing column. You can't change the key columns of a table.

data_type

• Data type of the new column, or new data type for an existing column.

• You can't change the data type of a generated column, or any columns referenced by the generated column.

• Changing the data type is not allowed on any columns referenced in a CHECK constraint. options_def

• The (allow_commit_timestamp=true) option allows insert and update operations to request that Spanner write the commit timestamp of the transaction into the column. For details, see Commit timestamps in GoogleSQL-dialect databases.

table_constraint

• New table constraint for the table.

constraint_name

• The name of a new or existing constraint.

ref_table

• The referenced table in a foreign key constraint.

ref_column

• The referenced column in a foreign key constraint.

DROP TABLE

Removes a table.

Syntax

DROP TABLE [ IF EXISTS ] table_name

Description

Use the DROP TABLE statement to remove a table from the database.

• DROP TABLE is not recoverable.

• You can't drop a table if there are indexes over it, or if there are any tables or indexes interleaved within it.

• A DROP TABLE statement automatically drops the foreign keys and foreign keys backing indexes of a table.

Parameters

IF EXISTS

• If a table of the specified name doesn't exist, then the DROP statement has no effect and no error is generated.

table_name

• The name of the table to drop.

INDEX statements

CREATE INDEX

Use the CREATE INDEX statement to define secondary indexes.

Syntax

CREATE [ UNIQUE ] [ NULL_FILTERED ] INDEX [ IF NOT EXISTS ] index_name
ON table_name ( key_part [, ...] ) [ storing_clause ] [ , interleave_clause ]

where index_name is:
    {a—z|A—Z}[{a—z|A—Z|0—9|_}+]

and key_part is:
    column_name [ { ASC | DESC } ]

and storing_clause is:
    STORING ( column_name [, ...] )

and interleave_clause is:
    INTERLEAVE IN table_name

Description

Spanner automatically indexes the primary key columns of each table.

You can use CREATE INDEX to create secondary indexes for other columns. Adding a secondary index on a column makes it more efficient to look up data in that column. For more details, see secondary indexes.

Note: Spanner has a hard size limit of 8 KB for the total size of a table key or an index key. To work around this limitation, you can create a stored generated column with expressions and an index column.

Parameters

UNIQUE

• Indicates that this secondary index enforces a UNIQUE constraint on the data being indexed. The UNIQUE constraint causes any transaction that would result in a duplicate index key to be rejected. See Unique Indexes for more information.

NULL_FILTERED

• Indicates that this secondary index does not index NULL values. See Indexing of NULL values for more information.

IF NOT EXISTS

• If an index already exists with the same name, then the CREATE statement has no effect and no error is generated.

index_name

• The name of the index to be created. For naming guidance, see Naming conventions.

table_name

• The name of the table to be indexed.

INTERLEAVE IN

• Defines a table to interleave the index in. If T is the table into which the index is interleaved, then:

  • T must be a parent of the table being indexed, and
  • The primary key of T must be the key prefix of the index.

  If the index key that you want to use for index operations matches the key of a table, you might want to interleave the index in that table if the row in the table should have a data locality relationship with the corresponding indexed rows.

  For example, if you want to index all rows of Songs for a particular row of Singers, your index keys would contain SingerId and SongName and your index would be a good candidate for interleaving in Singers if you frequently fetch information about a singer as you fetch that singer's songs from the index. The definition of SongsBySingerSongName in Creating a Secondary Index is an example of creating such an interleaved index.

  Like interleaved tables, entries in interleaved indexes are stored with the corresponding row of the parent table. See database splits for more details.

DESC

• Defines descending scan order for the corresponding index column. When scanning a table using an index column marked DESC, the scanned rows appear in the descending order with respect to this index column. If you don't specify a sort order, the default is ascending (ASC).

STORING

• Provides a mechanism for duplicating data from the table into one or more secondary indexes on that table. At the cost of extra storage, this can reduce read latency when looking up data using a secondary index, because it eliminates the need to retrieve data from the main table after having found the desired entries in the index. See STORING clause for an example.

ALTER INDEX

Use the ALTER INDEX statement to add additional columns or remove stored columns from the secondary indexes.

Syntax

ALTER INDEX index_name {ADD|DROP} STORED COLUMN column_name

Description

Add an additional column into an index or remove a column from an index.

Parameters

index_name

• The name of the index to alter.

column_name

• The name of the column to add into the index or to remove from the index.

DROP INDEX

Removes a secondary index.

Syntax

DROP INDEX [ IF EXISTS ] index_name

Description

Use the DROP INDEX statement to drop a secondary index.

Parameters

IF EXISTS

• If an index of the specified name doesn't exist, then the DROP statement has no effect and no error is generated.

index_name

• The name of the index to drop.

VIEW statements

CREATE VIEW and CREATE OR REPLACE VIEW

Use the CREATE VIEW or CREATE OR REPLACE VIEW statement to define a view.

Syntax

{ CREATE VIEW | CREATE OR REPLACE VIEW } view_name
SQL SECURITY { INVOKER | DEFINER }
AS query

Description

CREATE VIEW defines a new view in the current database. If a view named view_name exists, the CREATE VIEW statement fails.

CREATE OR REPLACE VIEW defines a new view in the current database. If a view named view_name exists, its definition is replaced. Use this statement to replace the security type of a view.

Parameters

view_name

• The name of the view to be created. For naming guidance, see Naming conventions.

SQL SECURITY

• The security type can be either INVOKER or DEFINER. Depending on the security type of the view, Spanner may or may not access check the objects referenced in the view against the database role of the principal who invoked the query. For more information, see About views.

AS query

• The query that defines the view content.

  • The query must specify a name for each item in the SELECT list.

  • The query cannot include query parameters.

  • GoogleSQL disregards any ORDER BY clause in this query that isn't paired with a LIMIT clause.

  See Query syntax for information on constructing a query.

DROP VIEW

Removes a view.

Syntax

DROP VIEW view_name

Description

Use the DROP VIEW statement to remove a view from the database.

Parameters

view_name

• The name of the view to drop.

CHANGE STREAM statements

CREATE CHANGE STREAM

Defines a new change stream.

Syntax

CREATE CHANGE STREAM change_stream_name
[ FOR { table_columns [, ... ] | ALL } ]
[ OPTIONS ( change_stream_option [, ... ] ) ]

where table_columns is:
    table_name [ ( [ column_name, ... ] ) ]

and change_stream_option is:
    { retention_period = 'duration' |
      value_capture_type = { 'OLD_AND_NEW_VALUES' | 'NEW_ROW' |'NEW_VALUES' } }

Description

CREATE CHANGE STREAM defines a new change stream in the current database. For more information, see Create a change stream.

Parameters

change_stream_name

• The name of the change stream to be created. For naming guidance, see Naming conventions.

FOR {table_columns[, ... ] | ALL }

• The FOR clause defines the tables and columns that are watched by the change stream.

• You can specify a list of table_columns to watch, where table_columns can be either of the following:

  • table_name: This watches the entire table, including all of the future columns when they are added to this table.

  • table_name ( [column_name, ... ] ): You can optionally specify a list of zero or more non-key columns following the table name. This watches only the primary key and the listed non-key columns of the table. With an empty list of non-key columns, table_name() watches only the primary key.

  Note: Primary key columns are always watched by the change stream. You only need to list the non-key columns. Listing any primary key columns is not allowed.

• ALL allows you to watch all tables and columns in the entire database, including all of the future tables and columns as soon as they are created.

• When the FOR clause is omitted, the change stream watches nothing.

OPTIONS (change_stream_option[, ... ] )

• The retention_period = 'duration' option allows you to specify how long a change stream retains its data. The duration must be in the range [1d, 7d] and can be specified in days, hours, minutes, or seconds. For example, the values 1d, 24h, 1440m, and 86400s are equivalent. The default is 1 day. For details, see Data retention.

• The value_capture_type option controls which values are captured for a changed row. It can be OLD_AND_NEW_VALUES (default), NEW_VALUES, or NEW_ROW. For details, see Value capture type.

ALTER CHANGE STREAM

Changes the definition of a change stream.

Syntax

ALTER CHANGE STREAM change_stream_name
    action

where action is:
    { SET FOR { table_columns [, ... ] | ALL } |
      DROP FOR ALL |
      SET OPTIONS ( change_stream_option [, ... ] ) }

and table_columns is:
    table_name [ ( [ column_name, ... ] ) ]

and change_stream_option is:
    { retention_period = { 'duration' | null } |
      value_capture_type = { 'OLD_AND_NEW_VALUES' | 'NEW_ROW' | 'NEW_VALUES' | null } }

Description

ALTER CHANGE STREAM changes the definition of an existing change stream. For more information, see Modify a change stream.

Parameters

change_stream_name

• The name of an existing change stream to alter.

SET FOR {table_columns[, ... ] | ALL }

• Sets a new FOR clause to modify what the change stream watches, using the same syntax as CREATE CHANGE STREAM.

DROP FOR ALL

• Suspends a change stream to watch nothing.

SET OPTIONS

• Sets options on the change stream (such as retention_period and value_capture_type), using the same syntax as CREATE CHANGE STREAM.

• Setting an option to null is equivalent to setting it to the default value.

DROP CHANGE STREAM

Removes a change stream.

Syntax

DROP CHANGE STREAM change_stream_name

Description

Use the DROP CHANGE STREAM statement to remove a change stream from the database and delete its data change records.

Parameters

change_stream_name

• The name of the change stream to drop.

ROLE statements

CREATE ROLE

Defines a new database role.

Syntax

CREATE ROLE database_role_name

Description

CREATE ROLE defines a new database role. Database roles are collections of fine-grained access control privileges. You can create only one role with this statement.

Parameters

database_role_name

• The name of the database role to create. The role name public and role names starting with spanner_ are reserved for system roles. See also Naming conventions.

Example

This example creates the database role hr_manager.

CREATE ROLE hr_manager

DROP ROLE

Drops a database role.

Syntax

DROP ROLE database_role_name

Description

DROP ROLE drops a database role. You can drop only one role with this statement.

You can't drop a database role if it has any privileges granted to it. All privileges granted to a database role must be revoked before the role can be dropped. You can drop a database role whether or not access to it is granted to IAM principals.

Dropping a role automatically revokes its membership in other roles and revokes the membership of its members.

You can't drop system roles.

Parameters

database_role_name

• The name of the database role to drop.

Example

This example drops the database role hr_manager.

DROP ROLE hr_manager

GRANT and REVOKE statements

GRANT

Syntax

GRANT { SELECT | INSERT | UPDATE | DELETE } ON TABLE table_list TO ROLE role_list

GRANT { SELECT | INSERT | UPDATE }(column_list) ON TABLE table_list TO ROLE role_list

GRANT SELECT ON CHANGE STREAM change_stream_list TO ROLE role_list

GRANT SELECT ON VIEW view_list TO ROLE role_list

GRANT EXECUTE ON TABLE FUNCTION function_list TO ROLE role_list

GRANT ROLE role_list TO ROLE role_list

where table_list is:
      table_name [, ...]

and column_list is:
    column_name [,...]

and view_list is:
    view_name [, ...]

and change_stream_list is:
    change_stream_name [, ...]

and function_list is:
    change_stream_read_function_name [, ...]

and role_list is:
    database_role_name [, ...]

Description

For fine-grained access control, grants privileges on one or more tables, views, change streams, or change stream read functions to database roles. Also grants database roles to other database roles to create a database role hierarchy with inheritance. When granting SELECT, INSERT, or UPDATE on a table, optionally grants privileges on only a subset of table columns.

Parameters

table_name

• The name of an existing table.

column_name

• The name of an existing column in the specified table.

view_name

• The name of an existing view.

change_stream_name

• The name of an existing change stream.

change_stream_read_function_name

• The name of an existing read function for a change stream. For more information, see Change stream read functions and query syntax.

database_role_name

• The name of an existing database role.

Notes and restrictions

• Identifiers for database objects named in the GRANT statement must use the case that was specified when the object was created. For example, if you created a table with a name that is in all lower case with a capitalized first letter, you must use that same case in the GRANT statement. Table-valued functions (TVFs) get automatically created with a prefix added to the change stream name, so ensure that you use the proper case for both the prefix and the change stream name. For more information about TVFs, see Change stream query syntax. created a table with a name that is in all lower case with a capitalized first letter, you must use that same case in the GRANT statement. For each change stream, PostgreSQL automatically creates a change stream read function with a name that consists of a prefix added to the change stream name, so ensure that you use the proper case for both the prefix and the change stream name. For more information about change stream read functions, see Change stream query syntax.

• When granting column-level privileges on multiple tables, each table must contain the named columns.

• If a table contains a column that is marked NOT NULL and has no default value, you can't insert into the table unless you have the INSERT privilege on that column.

• After granting SELECT on a change stream to a role, grant EXECUTE to that role on the read function for the change stream. For information about change stream read functions, see Change stream read functions and query syntax.

• Granting SELECT on a table doesn't grant SELECT on the change stream that tracks it. You must make a separate grant for the change stream.

Examples

The following example grants SELECT on the employees table to the hr_rep role. Grantees of the hr_rep role can read all columns of employees.

GRANT SELECT ON TABLE employees TO ROLE hr_rep;

The next example grants SELECT on a subset of columns of the contractors table to the hr_rep role. Grantees of the hr_rep role can read only the named columns.

GRANT SELECT(name, address, phone) ON TABLE contractors TO ROLE hr_rep;

The next example mixes table-level and column-level grants. hr_manager can read all table columns, but can update only the location column.

GRANT SELECT, UPDATE(location) ON TABLE employees TO ROLE hr_manager;

The next example makes column-level grants on two tables. Both tables must contain the name, level, and location columns.

GRANT SELECT(name, level, location), UPDATE(location) ON TABLE employees, contractors TO ROLE hr_manager;

The next example grants INSERT on a subset of columns of the employees table.

GRANT INSERT(name, cost_center, location, manager) ON TABLE employees TO ROLE hr_manager;

The next example grants the database role pii_access to the roles hr_manager and hr_director. The hr_manager and hr_director roles are members of pii_access and inherit the privileges that were granted to pii_access. For more information, see Database role hierarchies and inheritance.

GRANT ROLE pii_access TO ROLE hr_manager, hr_director;

REVOKE

Syntax

REVOKE { SELECT | INSERT | UPDATE | DELETE } ON TABLE table_list FROM ROLE role_list

REVOKE { SELECT | INSERT | UPDATE }(column_list) ON TABLE table_list FROM ROLE role_list

REVOKE SELECT ON VIEW view_list FROM ROLE role_list

REVOKE SELECT ON CHANGE STREAM change_stream_list FROM ROLE role_list

REVOKE EXECUTE ON TABLE FUNCTION function_list FROM ROLE role_list

REVOKE ROLE role_list FROM ROLE role_list

and table_list is:
    table_name [, ...]

and column_list is:
    column_name [,...]

and view_list is:
    view_name [, ...]

and change_stream_list is:
    change_stream_name [, ...]

and function_list is:
    change_stream_read_function_name [, ...]

and role_list is:
    database_role_name [, ...]

Description

For fine-grained access control, revokes privileges on one or more tables, views, change streams, or change stream read functions from database roles. Also revokes database roles from other database roles. When revoking SELECT, INSERT, or UPDATE on a table, optionally revokes privileges on only a subset of table columns.

Parameters

table_name

• The name of an existing table.

column_name

• The name of an existing column in the previously specified table.

view_name

• The name of an existing view.

change_stream_name

• The name of an existing change stream.

change_stream_read_function_name

• The name of an existing read function for a change stream. For more information, see Change stream read functions and query syntax.

database_role_name

• The name of an existing database role.

Notes and restrictions

• Identifiers for database objects named in the REVOKE statement must use the case that was specified when the object was created. For example, if you created a table with a name that is in all lower case with a capitalized first letter, you must use that same case in the REVOKE statement. For each change stream, PostgreSQL automatically creates a change stream read function with a name that consists of a prefix added to the change stream name, so ensure that you use the proper case for both the prefix and the change stream name. For more information about change stream read functions, see Change stream query syntax.

• When revoking column-level privileges on multiple tables, each table must contain the named columns.

• A REVOKE statement at the column level has no effect if privileges were granted at the table level.

• After revoking SELECT on a change stream from a role, revoke EXECUTE on the change stream's read function from that role.

• Revoking SELECT on a change stream doesn't revoke any privileges on the table that it tracks.

Examples

The following example revokes SELECT on the employees table from the role hr_rep.

REVOKE SELECT ON TABLE employees FROM ROLE hr_rep;

The next example revokes SELECT on a subset of columns of the contractors table from the role hr_rep.

REVOKE SELECT(name, address, phone) ON TABLE contractors FROM ROLE hr_rep;

The next example shows revoking both table-level and column-level privileges in a single statement.

REVOKE SELECT, UPDATE(location) ON TABLE employees FROM ROLE hr_manager;

The next example revokes column-level grants on two tables. Both tables must contain the name, level, and location columns.

REVOKE SELECT(name, level, location), UPDATE(location) ON TABLE employees, contractors FROM ROLE hr_manager;

The next example revokes INSERT on a subset of columns.

REVOKE INSERT(name, cost_center, location, manager) ON TABLE employees FROM ROLE hr_manager;

The following example revokes the database role pii_access from the hr_manager and hr_director database roles. The hr_manager and hr_director roles lose any privileges that they inherited from pii_access.

REVOKE ROLE pii_access FROM ROLE hr_manager, hr_director;

SEQUENCE statements

CREATE SEQUENCE

Syntax

CREATE SEQUENCE
    [ IF NOT EXISTS ] sequence_name
    [ OPTIONS ( sequence_options ) ]

Description

When you use a CREATE SEQUENCE statement, Spanner creates a schema object that you can poll for values using the GET_NEXT_SEQUENCE_VALUE function.

Parameters

IF NOT EXISTS

• If a sequence already exists with the same name, then the CREATE statement has no effect and no error is generated.

sequence_name

• The name of the sequence to create. For naming guidance, see Naming conventions.

OPTIONS (sequence_options)

• Use this clause to set an option on the specified sequence. Each sequence option uses a key=value pair, where key is the option name, and value is a literal. Multiple options are separated by commas. Options use the following syntax:

  OPTIONS (option_name = value [,...])
  

  A sequence accepts the following options:

  • The sequence_kind option accepts a STRING to indicate the type of sequence to use. At this time, bit_reversed_positive is the only valid type and it's a required option.
  • The skip_range_min and skip_range_max parameters cause the sequence to skip the numbers in this range when calling GET_NEXT_SEQUENCE_VALUE. These parameters are both integers that have a default value of NULL. The accepted values for skip_range_min is any value that is less than or equal to skip_range_max. The accepted values for skip_range_max is any value that is more than or equal to skip_range_min.
  • The start_with_counter option is a positive INT64 value that Spanner uses to set the next value for the internal sequence counter. For example, the next time that Spanner obtains a value from the bit-reversed sequence, it begins with start_with_counter. Spanner bit reverses this value before returning it to the client. The default value is 1.

Examples

# Create a positive bit-reversed sequence to use in a primary key.

CREATE SEQUENCE MySequence OPTIONS (
    sequence_kind='bit_reversed_positive',
    skip_range_min = 1,
    skip_range_max = 1000,
    start_with_counter = 50);

# Create a table that uses the sequence for a key column.
CREATE TABLE Singers (
  SingerId INT64 DEFAULT (GET_NEXT_SEQUENCE_VALUE(SEQUENCE MySequence)),
  FirstName  STRING(1024),
  LastName   STRING(1024),
  SingerInfo BYTES(MAX),
  BirthDate  DATE
) PRIMARY KEY (SingerId);

Use the following SQL to query information about sequences.

SELECT * FROM information_schema.sequences;
SELECT * FROM information_schema.sequence_options;

ALTER SEQUENCE

Syntax

ALTER SEQUENCE sequence_name
SET OPTIONS sequence_options

Description

ALTER SEQUENCE makes changes to the specified sequence schema object. Executing this statement doesn't affect values the sequence already generated. If the ALTER SEQUENCE statement doesn't include an option, the current value of the option remains the same.

Parameters

sequence_name

• The name of an existing sequence to alter. sequence_name is case sensitive. Don't include the path in the sequence_name.

SET OPTIONS (sequence_options)

• Use this clause to set an option on the specified sequence. Each sequence option uses a key=value pair, where key is the option name, and value is a literal. Multiple options are separated by commas. Options use the following syntax:

  SET OPTIONS (option_name = value [,...])
  

  This parameter offers the same options as CREATE SEQUENCE.

Examples

# Alter the sequence to include a skipped range. This is useful when you are
# migrating from a regular sequence with sequential data
ALTER SEQUENCE MySequence
SET OPTIONS (skip_range_min=1, skip_range_max=1234567);

DROP SEQUENCE

Syntax

DROP SEQUENCE [IF EXISTS] sequence_name

Description

DROP SEQUENCE drops a specific sequence. Spanner can't drop a sequence if its name appears in a sequence function that is used in a column default value or a view.

Parameters

sequence_name

• The name of the existing sequence to drop. IF EXISTS

• If a sequence of the specified name doesn't exist, then the DROP statement has no effect and no error is generated.

STATISTICS statements

ALTER STATISTICS

Changes the definition of a query optimizer statistics package.

Syntax

ALTER STATISTICS package_name
    action

where package_name is:
    {a—z}[{a—z|0—9|_|-}+]{a—z|0—9}

and action is:
    SET OPTIONS ( options_def )

and options_def is:
    { allow_gc = { true | false } }

Description

ALTER STATISTICS changes the definition of a query optimizer statistics package.

SET OPTIONS

• Use this clause to set an option on the specified statistics package.

Parameters

package_name

• The name of an existing query optimizer statistics package whose attributes are to be altered.

  To fetch existing statistics packages:

  SELECT s.package_name AS package_name, s.allow_gc AS allow_gc FROM INFORMATION_SCHEMA.SPANNER_STATISTICS s;

options_def

• The allow_gc = { true | false } option allows you to specify whether a given statistics package is garbage collected. A package must be set as allow_gc=false if it is used in a query hint. For more information, see Garbage collection of statistics packages.

ANALYZE

Start a new query optimizer statistics package construction.

Syntax

ANALYZE

Description

ANALYZE starts a new query optimizer statistics package construction.

MODEL statements

CREATE MODEL and CREATE OR REPLACE MODEL

Use the CREATE MODEL or CREATE OR REPLACE MODEL statement to define an ML model.

Note: Spanner Vertex AI integration supports only classifier, regression, and text ML models.

Syntax

{ CREATE MODEL | CREATE OR REPLACE MODEL | CREATE MODEL IF NOT EXISTS } model_name
[INPUT ( column_list ) OUTPUT ( column_list )]
REMOTE
[OPTIONS ( model_options )]

where column_list is:
   { column_name data_type [OPTIONS ( model_column_options )] [, ... ] }

and model_column_options is:
    {
      required = { true | false }
    }

and model_options is:
    {
      endpoint = '{endpoint_address}',
      endpoints = [ '{endpoint_address}' [, ...] ],
      default_batch_size = int64_value
    }

Description

CREATE MODEL registers a reference to the Vertex AI ML model in the current database. If a model named model_name already exists, the CREATE MODEL statement fails.

CREATE OR REPLACE MODEL registers a reference to the Vertex AI ML model in the current database. If a model named model_name already exists, its definition is replaced.

CREATE MODEL IF NOT EXISTS registers a reference to the Vertex AI ML model in the current database. If a model named model_name already exists, the CREATE MODEL IF NOT EXISTS statement does not have any effect and no error is generated.

As soon as the model reference is registered in a database, it can be used from queries that use the ML.Predict function.

Model registration doesn't result in copying a model from the Vertex AI to a database, but only in creation of a reference to this models' endpoint hosted in the Vertex AI. If the model's endpoint gets removed from the Vertex AI, Spanner queries referencing this model fail.

Model endpoint access control

To be able to access a registered Vertex AI model endpoint from Spanner, you need to grant access permission to Spanner's service agent account.

Spanner creates the service agent and grants the necessary permissions automatically when Spanner executes the first MODEL DDL statement. If both the Spanner database and the Vertex AI endpoint are in the same project, no additional setup is required.

If the Spanner service agent account doesn't exist for your Spanner project, create it by running the following command:

gcloud beta services identity create --service=spanner.googleapis.com --project={PROJECT}`

Follow the steps described in the following tutorial to grant the Spanner API Service Agent role to the Spanner service agent account service-{PROJECT}@gcp-sa-spanner.iam.gserviceaccount.com on your Vertex AI project.

Parameters

model_name

• The name of the model to be created. See Naming conventions.

INPUT ( column_list ) OUTPUT ( column_list )

• Lists of columns that define model inputs (that is, features) and outputs (that is, labels). The following types (used in the type field of column_list) are supported: BOOL, BYTES, FLOAT64, INT64, STRING, and ARRAY of listed types.

  • Map the model's input or output columns with 32-bit integer or floating-point number types to their 64-bit counterparts, which are INT64 and FLOAT64, respectively.

• If the Vertex AI endpoint has instance and prediction schemas, Spanner validates the provided INPUT and OUTPUT clauses against those remote schemas. You can also omit INPUT and OUTPUT clauses, letting Spanner automatically discover the endpoint schema.

• If the Vertex AI endpoint does not have instance and prediction schemas, INPUT and OUTPUT clauses must be provided. Spanner doesn't perform validation and mismatches result in runtime errors. We strongly recommend providing instance and prediction schemas, especially when using custom models.

model_column_options

• required lets you mark input or output columns as optional to match your Vertex AI schema.
  • Input columns cannot be declared optional if the instance field is required.
  • Optional input columns can be omitted in ML function calls.
  • Required input columns must be provided to ML function calls.
  • Output columns cannot be declared as required if the prediction field is optional.
  • Optional outputs columns can return NULL if the endpoint does not produce them.
  • Required outputs columns must be produced by the endpoint.

model_options

• endpoint is the address of the Vertex AI endpoint to connect to. Mutually exclusive with endpoints option. Supported formats:

  • //aiplatform.googleapis.com/projects/{project}/locations/{location}/endpoints/{endpoint}.
  • //aiplatform.googleapis.com/projects/{project}/locations/{location}/publishers/{publisher}/models/{endpoint}.
  • https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}/endpoints/{endpoint}.
  • https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}/publishers/{publisher}/models/{endpoint}.

• endpoints is a list of addresses of Vertex AI endpoints to connect to. Mutually exclusive with endpoint option. Prediction starts with the first endpoint on the list and fails over in the specified order. Endpoints can host different models as long as their schemas can be merged together:

  • Each column's name must use the same case across all endpoints
  • Each column's type must be the same across all endpoints.
  • Each input column is considered required if at least one endpoint requires it
  • Each output column is considered required only if all endpoints require it

• default_batch_size specifies the maximum number of rows per remote inference call. The value must be between 1 and 10. For models that do not support batching, you must set the value to 1. This default value can be overridden with per-query hints.

ALTER MODEL

Changes the definition of a model.

Note: Spanner Vertex AI integration supports only classifier, regression, and text ML models.

Syntax

ALTER MODEL [ IF EXISTS ] model_name
SET OPTIONS ( model_options )

where model_options is:
    {
      endpoint = '{endpoint_address}',
      endpoints = [ '{endpoint_address}' [, ...] ],
      default_batch_size = int64_value
    }

Description

ALTER MODEL changes the definition of an existing table.

Parameters

model_name

• The name of an existing model whose attributes are to be altered.

SET OPTIONS

• Sets options on the model, using the same syntax as CREATE MODEL.

• Setting an option to null is equivalent to setting it to the default value.

• Below are the list of options which can be updated:

  • endpoint is the address of the Vertex AI endpoint to connect to.
  • endpoints is a list of addresses of Vertex AI endpoints to connect to. Mutually exclusive with endpoint option.
  • default_batch_size specifies the maximum number of rows per remote inference call. The value must be between 1 and 10. For models that do not support batching, you must set the value to 1. This default value can be overridden with per-query hints.

DROP MODEL

Removes a model.

Syntax

DROP MODEL [ IF EXISTS ] model_name

Description

Use the DROP MODEL statement to remove a model definition from the database. Unless the IF EXISTS clause is specified, the statement fails if the model doesn't exist.

After you delete a model definition, all SQL queries referencing the deleted model fail. Dropping a model definition does not affect the underlying the Vertex AI endpoint that this model is attached to.

Parameters

model_name

• The name of the model to drop.