Skip to content

sql-server-linux-setup-machine-learning.md

Latest commit

 

History

History
680 lines (469 loc) · 25.1 KB

sql-server-linux-setup-machine-learning.md

File metadata and controls

680 lines (469 loc) · 25.1 KB
title Install SQL Server Machine Learning Services (Python, R) on Linux
description Learn how to install SQL Server Machine Learning Services (Python and R) on Linux: Red Hat, Ubuntu, and SUSE.
author cawrites
ms.author chadam
ms.reviewer vanto
manager cgronlun
ms.date 01/22/2020
ms.topic conceptual
ms.prod sql
ms.technology machine-learning
monikerRange >=sql-server-ver15||>=sql-server-linux-ver15||=sqlallproducts-allversions

Install SQL Server Machine Learning Services (Python and R) on Linux

[!INCLUDEappliesto-ss-xxxx-xxxx-xxx-md-linuxonly]

This article guides you in the installation of SQL Server Machine Learning Services on Linux.

What are Machine Learning Services

Machine Learning Services is a feature add-on to the database engine.

Install and configure the SQL Server database engine first so that you can resolve any issues before adding more components.

Prepare Environment

The following Linux distributions are supported:

  • Red Hat Enterprise Linux (RHEL)
  • SUSE Linux Enterprise Server (SLES)
  • Ubuntu

Check the SQL Server Linux repositories for the Python and R extensions. If you already configured source repositories for the database engine install, you can run the mssql-mlservices package install commands using the same repo registration.

Linux containers support Machine Learning Services. No pre-built containers available with Machine Learning Services, but you can create one from the SQL Server containers using an example template available on GitHub.

Machine Learning Services is installed by default on SQL Server Big Data Clusters. For more information, see Use Machine Learning Services (Python and R) on Big Data Clusters.

Python and R scripts can be executed in-database using Machine Learning Services.

Uninstall preview release

  • Uninstall the preview release. (Community Technical Preview (CTP) or Release Candidate)
  • No support for side-by-side installation of multiple versions, and the package list has changed over the last several preview (CTP/RC) releases.

The following files indicate an existing installation: setnetbr, checkinstallextensibility.sh, launchpadd.

ls /opt/microsoft/mssql/bin

1. Uninstall CTP/RC packages

Uninstall at the lowest package level. Any upstream package dependent on a lower-level package is automatically uninstalled.

  • For Python integration, remove mssql-mlservices-python
  • For R integration, remove microsoft-r-open*

Commands for removing packages appear in the following table.

Platform Package removal command(s)
Red Hat sudo yum remove microsoft-r-open-mro-3.4.4
sudo yum remove msssql-mlservices-python
SUSE sudo zypper remove microsoft-r-open-mro-3.4.4
sudo zypper remove msssql-mlservices-python
Ubuntu sudo apt-get remove microsoft-r-open-mro-3.4.4
sudo apt-get remove msssql-mlservices-python

Note

Microsoft R Open 3.4.4 is composed of two packages, depending on which CTP release you previously installed. (The foreachiterators package was combined into the main mro package in CTP 2.2.) If any of these packages remain after removing microsoft-r-open-mro-3.4.4, you should remove them individually.

microsoft-r-open-foreachiterators-3.4.4
microsoft-r-open-mkl-3.4.4
microsoft-r-open-mro-3.4.4

2. Install

Install at the highest package level using the instructions in this article for your operating system.

Installation options highest package level:

  • Example 1 - Full installation for the full set of packages.
  • Example 2 - Minimal installation for the least number of packages required for an installation.

  1. The prerequisite For R integration; start with MRO.

  2. Run install commands using the package managers and syntax for your operating system:

Prerequisites

Microsoft R Open (MRO) installation

Microsoft's base distribution of R is a prerequisite for using RevoScaleR, MicrosoftML, and other R packages installed with Machine Learning Services.

The required version is MRO 3.5.2.

Choose from the following two approaches to install MRO:

  • Download the MRO tarball from MRAN, unpack it, and run its install.sh script. You can follow the installation instructions on MRAN if you want this approach.

  • Register the packages.microsoft.com repo as described below to install the MRO distribution: microsoft-r-open-mro and microsoft-r-open-mkl.

The following commands register the repository providing MRO. Post-registration, the commands for installing other R packages, such as mssql-mlservices-mml-r, will automatically include MRO as a package dependency.

MRO on Ubuntu

# Install as root
sudo su

# Optionally, if your system does not have the https apt transport option
apt-get install apt-transport-https

# Set the location of the package repo the "prod" directory containing the distribution.
# This example specifies 16.04. Replace with 14.04 if you want that version
wget https://packages.microsoft.com/config/ubuntu/16.04/packages-microsoft-prod.deb

# Register the repo
dpkg -i packages-microsoft-prod.deb

# Update packages on your system (required), including MRO installation
sudo apt-get update

MRO on Red Hat

# Import the Microsoft repository key
sudo rpm --import https://packages.microsoft.com/keys/microsoft.asc


# Set the location of the package repo at the "prod" directory
# The following command is for version 7.x
# For 6.x, replace 7 with 6 to get that version
rpm -Uvh https://packages.microsoft.com/config/rhel/7/packages-microsoft-prod.rpm

# Update packages on your system (optional)
yum update

MRO on SUSE

# Install as root
sudo su

# Set the location of the package repo at the "prod" directory containing the distribution
# This example is for SLES12, the only supported version of SUSE in Machine Learning Server
zypper ar -f https://packages.microsoft.com/sles/12/prod packages-microsoft-com

# Update packages on your system (optional)
zypper update

Package list

Available installation Packages:

Package name Applies-to Description
mssql-server-extensibility All Extensibility framework used to run Python and R.
microsoft-openmpi Python, R Message passing interface used by the Rev* libraries for parallelization on Linux.
mssql-mlservices-python Python Open-source distribution of Anaconda and Python.
mssql-mlservices-mlm-py Python Full install. Provides revoscalepy, microsoftml, pre-trained models for image featurization and text sentiment analysis.
mssql-mlservices-packages-py Python Minimum install. Provides revoscalepy and microsoftml.
Excludes pre-trained models.
microsoft-r-open* R Open-source distribution of R, composed of three packages.
mssql-mlservices-mlm-r R Full install. Provides: RevoScaleR, MicrosoftML, sqlRUtils, olapR, pre-trained models for image featurization and text sentiment analysis.
mssql-mlservices-packages-r R Minimum install. Provides RevoScaleR, sqlRUtils, MicrosoftML, olapR.
Excludes pre-trained models.

RedHat commands

Installation Options for Python and R:

  • Install language support based on your requirements (single or multiple languages).
  • The full installation provides all available features the including pre-trained machine learning models.
  • The minimal installation excludes the models but still has all of the functionality.

Tip

If possible, run yum clean all to refresh packages on the system prior to installation.

Example 1 - Full installation

Includes:

  • open-source Python
  • open-source R
  • extensibility framework
  • microsoft-openmpi
  • extensions (Python, R)
  • machine learning libraries
  • pre-trained models for Python and R
# Install as root or sudo
# Add everything (all R, Python)
# Be sure to include -9.4.7* in mlsservices package names
sudo yum install mssql-mlservices-mlm-py-9.4.7*
sudo yum install mssql-mlservices-mlm-r-9.4.7*

Example 2 - Minimum installation

Includes:

  • open-source Python
  • open-source R
  • extensibility framework
  • microsoft-openmpi
  • core Revo* libraries
  • machine learning libraries
# Install as root or sudo
# Minimum install of R, Python extensions
# Be sure to include -9.4.6* in mlsservices package names
sudo yum install mssql-mlservices-packages-py-9.4.7*
sudo yum install mssql-mlservices-packages-r-9.4.7*

Ubuntu commands

Installation Options for Python and R:

  • Install language support based on your requirements (single or multiple languages).
  • The full installation provides all available features the including pre-trained machine learning models.
  • The minimal installation excludes the models but still has all of the functionality.

Tip

If possible, run apt-get update to refresh packages on the system prior to installation. Additionally, some docker images of Ubuntu might not have the https apt transport option. To install it, use apt-get install apt-transport-https.

Example 1 - Full installation

Includes:

  • open-source Python
  • opsn-source R
  • extensibility framework
  • microsoft-openmpi
  • Python extensions
  • R extensions
  • machine learning libraries
  • pre-trained models for Python and R
# Install as root or sudo
# Add everything (all R, Python)
# There is no asterisk in this full install
sudo apt-get install mssql-mlservices-mlm-py 
sudo apt-get install mssql-mlservices-mlm-r

Example 2 - Minimum installation

Includes:

  • open-source Python
  • open-source R
  • extensibility framework
  • microsoft-openmpi
  • core Revo* libraries
  • machine learning libraries
# Install as root or sudo
# Minimum install of R, Python
# No aasterisk
sudo apt-get install mssql-mlservices-packages-py
sudo apt-get install mssql-mlservices-packages-r

SUSE commands

Installation Options for Python and R:

  • Install language support based on your requirements (single or multiple languages).
  • The full installation provides all available features the including pre-trained machine learning models.
  • The minimal installation excludes the models but still has all of the functionality.

Example 1 - Full installation

Includes:

  • open-source Python
  • open-source R
  • extensibility framework
  • microsoft-openmpi
  • extensions for Python and R
  • machine learning libraries
  • pre-trained models for Python and R
# Install as root or sudo
# Add everything (all R, Python)
# Be sure to include -9.4.7* in mlsservices package names
sudo zypper install mssql-mlservices-mlm-py-9.4.7*
sudo zypper install mssql-mlservices-mlm-r-9.4.7*

Example 2 - Minimum installation

Includes:

  • open-source Python
  • open-source R
  • extensibility framework
  • microsoft-openmpi
  • core Revo* libraries
  • machine learning libraries
# Install as root or sudo
# Minimum install of R, Python extensions
# Be sure to include -9.4.6* in mlsservices package names
sudo zypper install mssql-mlservices-packages-py-9.4.7*
sudo zypper install mssql-mlservices-packages-r-9.4.7*

Post-install config (required)

Additional configuration is primarily through the mssql-conf tool.

  1. Add the mssql user account used to run the SQL Server service.

    sudo /opt/mssql/bin/mssql-conf setup

  2. Accept the licensing agreements for open-source Python and R extensions. Use the following command:

    # Run as SUDO or root
# Use set + EULA 
sudo /opt/mssql/bin/mssql-conf set EULA accepteulaml Y

  3. Setup detects the mssql-mlservices packages and prompts for EULA acceptance (if not previously accepted) when mssql-conf setup is run. For more information about EULA parameters, see Configure SQL Server with the mssql-conf tool.

  4. Enable outbound network access. Outbound network access is disabled by default. To enable outbound requests, set the "outboundnetworkaccess" Boolean property using the mssql-conf tool. For more information, see Configure SQL Server on Linux with mssql-conf.

    # Run as SUDO or root
# Enable outbound requests over the network
sudo /opt/mssql/bin/mssql-conf set extensibility outboundnetworkaccess 1

  5. For R feature integration only, set the MKL_CBWR environment variable to ensure consistent output from Intel Math Kernel Library (MKL) calculations.

    • Edit or create a file named.bash_profile in your user home directory, adding the line export MKL_CBWR="AUTO" to the file.

    • Execute this file by typing source.bash_profile at a bash command prompt.

  6. Restart the SQL Server Launchpad service and the database engine instance to read the updated values from the INI file. A notification is message displayed when an extensibility-related setting is modified.

    systemctl restart mssql-launchpadd

systemctl restart mssql-server.service

  7. Enable external script execution using Azure Data Studio or another tool like SQL Server Management Studio (Windows only) that runs Transact-SQL.

    EXEC sp_configure 'external scripts enabled', 1 
RECONFIGURE WITH OVERRIDE

  8. Restart the Launchpad service again.

Validate installation

R libraries (MicrosoftML, RevoScaleR, and others) can be found at /opt/mssql/mlservices/libraries/RServer.

Python libraries (microsoftml and revoscalepy) can be found at /opt/mssql/mlservices/libraries/PythonServer.

To validate installation:

  • Run a T-SQL script that executes a system stored procedure invoking Python or R using a query tool.

For Windows use:

  • Azure Data Studio
  • SQL Server Management Studio or PowerShell

If you have a Windows computer with these tools, use it to connect to your Linux installation of the database engine.

Execute the following SQL command to test R execution in SQL Server. Errors? Try a service restart, sudo systemctl restart mssql-server.service.

EXEC sp_execute_external_script   
@language =N'R', 
@script=N' 
OutputDataSet <- InputDataSet', 
@input_data_1 =N'SELECT 1 AS hello' 
WITH RESULT SETS (([hello] int not null)); 
GO

Execute the following SQL command to test Python execution in SQL Server.

EXEC sp_execute_external_script  
@language =N'Python', 
@script=N' 
OutputDataSet = InputDataSet; 
', 
@input_data_1 =N'SELECT 1 AS hello' 
WITH RESULT SETS (([hello] int not null)); 
GO

Chained "combo" install

  1. Completed the following steps:
  • Install and configure the database engine and Machine Learning Services.
  • Append Python or R packages and parameters on a command that installs the database engine.
  • For R integration, install Microsoft R Open as a prerequisite. Skip this step if not needed.
  1. Create a command line that includes the database engine, plus language extension features.

You can add a single feature, such as Python integration, to a database engine install.

sudo yum install -y mssql-server mssql-mlservices-packages-r-9.4.7*

Or, add both extensions (Python, R).

sudo yum install -y mssql-server mssql-mlservices-packages-r-9.4.7* mssql-mlservices-packages-py-9.4.7*
  1. Accept license agreements and complete the post-install configuration. Use the mssql-conf tool for this task.
sudo /opt/mssql/bin/mssql-conf setup

Accept prompt for the license agreement for the database engine, choose an edition, and set the administrator password. Accept the license agreement for Machine Learning Services.

  1. Restart the service, if prompted to do so.
sudo systemctl restart mssql-server.service

Unattended installation

Using the unattended install for the Database Engine, add the packages for mssql-mlservices and EULAs.

Use one of the mlservices-specific EULA parameters for the open-source R and Python distributions:

sudo /opt/mssql/bin/mssql-conf setup accept-eula-ml

The complete EULA is documented at Configure SQL Server on Linux with the mssql-conf tool.

Offline installation

Follow the Offline installation instructions for steps on installing the packages. Download specific packages using the package list below.

Tip

Several of the package management tools provide commands that can help you determine package dependencies. For yum, use sudo yum deplist [package]. For Ubuntu, use sudo apt-get install --reinstall --download-only [package name] followed by dpkg -I [package name].deb.

Download site

Download packages from https://packages.microsoft.com/. All of the mlservices packages for Python and R are colocated with database engine package. Base version for the mlservices packages is 9.4.6. Recall that the microsoft-r-open packages are in a different repository.

RHEL/7 paths
mssql/mlservices packages https://packages.microsoft.com/rhel/7/mssql-server-preview/
microsoft-r-open packages https://packages.microsoft.com/rhel/7/prod/

Ubuntu/16.04 paths
mssql/mlservices packages https://packages.microsoft.com/ubuntu/16.04/mssql-server-preview/pool/main/m/
microsoft-r-open packages https://packages.microsoft.com/ubuntu/16.04/prod/pool/main/m/

SLES/12 paths
mssql/mlservices packages https://packages.microsoft.com/sles/12/mssql-server-preview/
microsoft-r-open packages https://packages.microsoft.com/sles/12/prod/

Package list

Select extensions you want to use and download the packages necessary for a specific language. The filenames include platform information in the suffix.

File List:

# Core packages 
mssql-server-15.0.1000
mssql-server-extensibility-15.0.1000

# R
microsoft-openmpi-3.0.0
microsoft-r-open-mkl-3.5.2
microsoft-r-open-mro-3.5.2
mssql-mlservices-packages-r-9.4.7.64
mssql-mlservices-mlm-r-9.4.7.64


# Python
microsoft-openmpi-3.0.0
mssql-mlservices-python-9.4.7.64
mssql-mlservices-packages-py-9.4.7.64
mssql-mlservices-mlm-py-9.4.7.64

Add more Python/R packages

You can install other R and Python packages and use them in script that executes on SQL Server 2019.

Python packages

  1. Install a Python package called httpie using pip.

    # sudo /opt/mssql/mlservices/bin/python/python -m pip install httpie

  2. Import the Python package in sp_execute_external_script.

    EXEC sp_execute_external_script  
@language = N'Python',  
@script = N'import httpie'

R packages

  1. Start an R session.

    # sudo /opt/mssql/mlservices/bin/R/R

  2. Install an R package called glue to test package installation.

    # install.packages("glue",lib="/opt/mssql/mlservices/libraries/RServer")

    Optionally, install an R package from the command line

    # sudo /opt/mssql/mlservices/bin/R/R CMD INSTALL -l /opt/mssql/mlservices/libraries/RServer glue_1.1.1.tar.gz

  3. Import the R package in sp_execute_external_script.

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

Run in a container

Build and run SQL Server Machine Learning Services in a Docker container. For more information, see Configure SQL Server container images on Docker.

Prerequisites

Clone the mssql-docker repository

  1. Open a Bash terminal on Linux or Mac, or open a WSL terminal on Windows.

  2. Create a local directory to hold a local copy of the mssql-docker repository.

  3. Run the git clone command to clone the mssql-docker repository:

    git clone https://github.com/microsoft/mssql-docker mssql-docker

Build a SQL Server Linux container image with Machine Learning Services

  1. Change the directory to the mssql-mlservices directory:

    cd mssql-docker/linux/preview/examples/mssql-mlservices

  2. Run the build.sh script:

    ./build.sh

    [!NOTE] To build the Docker image, you must install packages that are several GBs in size. The script may take up to 20 minutes to finish running, depending on network bandwidth.

Run the SQL Server Linux container image with Machine Learning Services

  1. Set your environment variables before running the container. Set the PATH_TO_MSSQL environment variable to a host directory:

     export MSSQL_PID='Developer'
 export ACCEPT_EULA='Y'
 export ACCEPT_EULA_ML='Y'
 export PATH_TO_MSSQL='/home/mssql/'

  2. Run the run.sh script:

    ./run.sh

    This command creates a SQL Server container with Machine Learning Services, using the Developer edition (default). SQL Server port 1433 is exposed on the host as port 1401.

    [!NOTE] The process for running production SQL Server editions in containers is slightly different. For more information, see Configure SQL Server container images on Docker. If you use the same container names and ports, the rest of this walk-through still works with production containers.

  3. To view your Docker containers, run the docker ps command:

    sudo docker ps -a

  • Up - SQL Server is running in the container and listening on the port specified in the PORTS column.

  • Exited - see the Troubleshooting section of the configuration guide.

    $ sudo docker ps -a

    Output:

    CONTAINER ID        IMAGE                          COMMAND                  CREATED             STATUS              PORTS                    NAMES
941e1bdf8e1d        mcr.microsoft.com/mssql/server/mssql-server-linux   "/bin/sh -c /opt/m..."   About an hour ago   Up About an hour     0.0.0.0:1401->1433/tcp   sql1

Next steps

R developers can get started with some simple examples, and learn the basics of how R works with SQL Server. For your next step, see the following links:

Python developers can learn how to use Python with SQL Server by following these tutorials:

To view examples of machine learning that are based on real-world scenarios, see Machine learning tutorials.