CREATE EXTERNAL LIBRARY (Transact-SQL)

Applies to: SQL Server 2017 (14.x) and later Azure SQL Managed Instance

Uploads R, Python, or Java package files to a database from the specified byte stream or file path. This statement serves as a generic mechanism for the database administrator to upload artifacts needed for any new external language runtimes and OS platforms supported by SQL Server.

Note

In SQL Server 2017, R language and Windows platform are supported. R, Python, and external languages on the Windows and Linux platforms are supported in SQL Server 2019 and later.

Uploads R or Python package files to a database from the specified byte stream or file path. This statement serves as a generic mechanism for the database administrator to upload artifacts needed.

Note

In Azure SQL Managed Instance, you can use sqlmlutils to install a library. For details, see Install Python packages with sqlmlutils and Install new R packages with sqlmlutils.

Syntax for SQL Server 2019

CREATE EXTERNAL LIBRARY library_name  
[ AUTHORIZATION owner_name ]  
FROM <file_spec> [ ,...2 ]  
WITH ( LANGUAGE = <language> )  
[ ; ]  

<file_spec> ::=  
{  
    (CONTENT = { <client_library_specifier> | <library_bits> }  
    [, PLATFORM = <platform> ])  
}  

<client_library_specifier> :: = 
{
    '[file_path\]manifest_file_name'  
} 

<library_bits> :: =  
{ 
      varbinary_literal 
    | varbinary_expression 
}

<platform> :: = 
{
      WINDOWS
    | LINUX
}

<language> :: = 
{
      'R'
    | 'Python'
    | <external_language>
}

Syntax for SQL Server 2017

CREATE EXTERNAL LIBRARY library_name  
[ AUTHORIZATION owner_name ]  
FROM <file_spec> [ ,...2 ]  
WITH ( LANGUAGE = 'R' )  
[ ; ]  

<file_spec> ::=  
{  
    (CONTENT = { <client_library_specifier> | <library_bits> })  
}  

<client_library_specifier> :: = 
{
    '[file_path\]manifest_file_name'
} 

<library_bits> :: =  
{ 
      varbinary_literal 
    | varbinary_expression 
}

Syntax for Azure SQL Managed Instance

CREATE EXTERNAL LIBRARY library_name  
[ AUTHORIZATION owner_name ]  
FROM <file_spec> [ ,...2 ]  
WITH ( LANGUAGE = <language> )  
[ ; ]  

<file_spec> ::=  
{  
    (CONTENT = <library_bits>)  
}  

<library_bits> :: =  
{ 
      varbinary_literal 
    | varbinary_expression 
}

<language> :: = 
{
      'R'
    | 'Python'
}

Arguments

library_name

Libraries uploaded to the instance can be either public or private. If the library is created by a member of dbo, the library is public and can be shared with all users. Otherwise, the library is private to that user only.

Library names must be unique within the context of a specific user or owner. For example, two users RUser1 and RUser2 can both individually and separately upload the R library ggplot2. However, if RUser1 wanted to upload a newer version of ggplot2, the second instance must be named differently or must replace the existing library.

Library names cannot be arbitrarily assigned; the library name should be the same as the name required to load the library in the external script.

owner_name

Specifies the name of the user or role that owns the external library. If not specified, ownership is given to the current user.

The libraries owned by database owner are considered global to the database and runtime. In other words, database owners can create libraries that contain a common set of libraries or packages that are shared by many users. When an external library is created by a user other than the dbo user, the external library is private to that user only.

When the user RUser1 executes an external script, the value of libPath can contain multiple paths. The first path is always the path to the shared library created by the database owner. The second part of libPath specifies the path containing packages uploaded individually by RUser1.

file_spec

Specifies the content of the package for a specific platform. Only one file artifact per platform is supported.

The file can be specified in the form of a local path, or network path.

When attempting to access the file specified in <client_library_specifier>, SQL Server impersonates the security context of the current Windows login. If <client_library_specifier> specifies a network location (UNC path), the impersonation of the current login is not carried forward to the network location because of delegation limitations. In this case, access is made using the security context of the SQL Server service account. For more information, see Credentials (Database Engine).

Optionally, an OS platform for the file can be specified. Only one file artifact or content is permitted for each OS platform for a specific language or runtime.

library_bits

Specifies the content of the package as a hex literal, similar to assemblies.

This option is useful if you need to create a library or alter an existing library (and have the required permissions to do so), but the file system on the server is restricted and you cannot copy the library files to a location that the server can access.

PLATFORM

Specifies the platform for the content of the library. The value defaults to the host platform on which SQL Server is running. Therefore, the user doesn't have to specify the value. It is required in case where multiple platforms are supported, or the user needs to specify a different platform. In SQL Server 2019, Windows and Linux are the supported platforms.

LANGUAGE = 'R'

Specifies the language of the package. R is supported in SQL Server 2017.

language

Specifies the language of the package. The value can be R or Python in Azure SQL Managed Instance.

language

Specifies the language of the package. The value can be R, Python, or the name of an external language (see CREATE EXTERNAL LANGUAGE).

Remarks

For the R language, when using a file, packages must be prepared in the form of zipped archive files with the .ZIP extension for Windows. In SQL Server 2017, only the Windows platform is supported.

For the R language, when using a file, packages must be prepared in the form of zipped archive files with the .ZIP extension.

For the Python language, the package in a .whl or .zip file must be prepared in the form of a zipped archive file. If the package already is a .zip file, it must be included in a new .zip file. Uploading a package as .whl or .zip file directly is currently not supported.

The CREATE EXTERNAL LIBRARY statement uploads the library bits to the database. The library is installed when a user runs an external script using sp_execute_external_script and calls the package or library.

Libraries uploaded to the instance can be either public or private. If the library is created by a member of dbo, the library is public and can be shared with all users. Otherwise, the library is private to that user only.

A number of packages, referred to as system packages, are pre-installed in a SQL instance. System packages cannot be added, updated, or removed by the user.

Permissions

Requires the CREATE EXTERNAL LIBRARY permission. By default, any user who has dbo who is a member of the db_owner role has permissions to create an external library. For all other users, you must explicitly give them permission using a GRANT statement, specifying CREATE EXTERNAL LIBRARY as the privilege.

In SQL Server 2019, in addition to 'CREATE EXTERNAL LIBRARY' permission, the user also needs references permission on an external language in order to create external libraries for that external language.

GRANT REFERENCES ON EXTERNAL LANGUAGE::Java to user
GRANT CREATE EXTERNAL LIBRARY to user

To modify any library requires the separate permission, ALTER ANY EXTERNAL LIBRARY.

To create an external library by using a file path, the user must be a Windows authenticated login or a member of the sysadmin fixed server role.

Examples

Add an external library to a database

The following example adds an external library called customPackage to a database.

CREATE EXTERNAL LIBRARY customPackage
FROM (CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL14.MSSQLSERVER\customPackage.zip') WITH (LANGUAGE = 'R');

After the library has been successfully uploaded to the instance, a user executes the sp_execute_external_script procedure, to install the library.

EXEC sp_execute_external_script 
@language =N'R', 
@script=N'library(customPackage)'

For the Python language in SQL Server 2019, the example also works by replacing 'R' with 'Python'.

Installing packages with dependencies

If the package you want to install has any dependencies, it is critical that you analyze both first-level and second-level dependencies, and ensure that all required packages are available before you try to install the target package.

For example, assume you want to install a new package, packageA:

  • packageA has a dependency on packageB
  • packageB has a dependency on packageC

To succeed in installing packageA, you must create libraries for packageB and packageC at the same time that you add packageA to SQL Server. Be sure to check the required package versions as well.

In practice, package dependencies for popular packages are usually much more complicated than this simple example. For example, ggplot2 might require over 30 packages, and those packages might require additional packages that are not available on the server. Any missing package or wrong package version can cause installation to fail.

Because it can be difficult to determine all dependencies just from looking at the package manifest, we recommend that you use a package such as miniCRAN to identify all packages that might be required to complete installation successfully.

  • Upload the target package and its dependencies. All files must be in a folder that is accessible to the server.

    CREATE EXTERNAL LIBRARY packageA 
    FROM (CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL14.MSSQLSERVER\packageA.zip') 
    WITH (LANGUAGE = 'R'); 
    GO
    
    CREATE EXTERNAL LIBRARY packageB FROM (CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL14.MSSQLSERVER\packageB.zip') 
    WITH (LANGUAGE = 'R');
    GO
    
    CREATE EXTERNAL LIBRARY packageC FROM (CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL14.MSSQLSERVER\packageC.zip') 
    WITH (LANGUAGE = 'R');
    GO
    
  • Install the required packages first.

    If a required package has already been uploaded to the instance, you need not add it again. Just be sure to check whether the existing package is the correct version.

    The required packages packageC and packageB are installed, in the correct order, when sp_execute_external_script is first run to install package packageA.

    However, if any required package is not available, installation of the target package packageA fails.

    EXEC sp_execute_external_script 
    @language =N'R', 
    @script=N'
    # load the desired package packageA
    library(packageA)
    '
    

For the Python language in SQL Server 2019, the example also works by replacing 'R' with 'Python'.

Create a library from a byte stream

If you do not have the ability to save the package files in a location on the server, you can pass the package contents in a variable. The following example creates a library by passing the bits as a hexadecimal literal.

CREATE EXTERNAL LIBRARY customLibrary FROM (CONTENT = 0xABC123...) WITH (LANGUAGE = 'R');

For the Python language in SQL Server 2019, the example also works by replacing 'R' with 'Python'.

Note

This code sample only demonstrates the syntax; the binary value in CONTENT = has been truncated for readability and does not create a working library. The actual contents of the binary variable would be much longer.

Change an existing package library

The ALTER EXTERNAL LIBRARY DDL statement can be used to add new library content or modify existing library content. To modify an existing library requires the ALTER ANY EXTERNAL LIBRARY permission.

For more information, see ALTER EXTERNAL LIBRARY.

Add a Java .jar file to a database

The following example adds an external jar file called customJar to a database.

CREATE EXTERNAL LIBRARY customJar
FROM (CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL15.MSSQLSERVER\customJar.jar') 
WITH (LANGUAGE = 'Java');

After the library has been successfully uploaded to the instance, a user executes the sp_execute_external_script procedure, to install the library.

EXEC sp_execute_external_script
    @language = N'Java'
    , @script = N'customJar.MyCLass.myMethod'
    , @input_data_1 = N'SELECT * FROM dbo.MyTable'
WITH RESULT SETS ((column1 int))

Add an external package for both Windows and Linux

You can specify up to two <file_spec>, one for Windows and one for Linux.

CREATE EXTERNAL LIBRARY lazyeval 
FROM (CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL15.MSSQLSERVER\packageA.zip', PLATFORM = WINDOWS),
(CONTENT = 'C:\Program Files\Microsoft SQL Server\MSSQL15.MSSQLSERVER\packageA.tar.gz', PLATFORM = LINUX)
WITH (LANGUAGE = 'R')

When you use sp_execute_external_script to install the package, depending on the platform the SQL Server instance is running on, the library content for that platform will be used.

EXECUTE sp_execute_external_script 
    @LANGUAGE = N'R',
    @SCRIPT = N'
library(packageA)'

See also

ALTER EXTERNAL LIBRARY (Transact-SQL)
DROP EXTERNAL LIBRARY (Transact-SQL)
sys.external_library_files
sys.external_libraries