Επεξεργασία

Κοινή χρήση μέσω

Data platforms supported on the Data Science Virtual Machine

  • Άρθρο

With a Data Science Virtual Machine (DSVM), you can build your analytics resources against a wide range of data platforms. In addition to interfaces to remote data platforms, the DSVM provides a local instance for rapid development and prototyping.

The DSVM supports these data platform tools:

SQL Server Developer Edition

Category Value
What is it? A local relational database instance
Supported DSVM editions Windows 2019, Linux (SQL Server 2019)
Typical uses
  • Rapid local development, with a smaller dataset
  • Run In-database R
Links to samples
  • A small sample of a New York City dataset is loaded into the SQL database:
    nyctaxi
  • Find a Jupyter sample that shows Microsoft Machine Learning Server and in-database analytics at:
    ~notebooks/SQL_R_Services_End_to_End_Tutorial.ipynb
Related tools on the DSVM
  • SQL Server Management Studio
  • ODBC/JDBC drivers
  • pyodbc, RODBC

Note

SQL Server Developer Edition can be used only for development and test purposes. You need a license or one of the SQL Server VMs to run it in production.

Note

Support for Machine Learning Server Standalone ended on July 1, 2021. We will remove it from the DSVM images after June 30. Existing deployments will continue to have access to the software but due to the reached support end date, support for it ended after July 1, 2021.

Note

We will remove SQL Server Developer Edition from DSVM images by end of November, 2021. Existing deployments will continue to have SQL Server Developer Edition installed. In new deployemnts, if you would like to have access to the SQL Server Developer Edition, you can install and use the SQL Server Developer Edition via Docker support. Visit Quickstart: Run SQL Server container images with Docker for more information.

Windows

Setup

The database server is already preconfigured, and the Windows services related to SQL Server (for example, SQL Server (MSSQLSERVER)) are set to run automatically. The only manual step involves enabling in-database analytics through use of Microsoft Machine Learning Server. Run the following command to enable analytics as a one-time action in SQL Server Management Studio (SSMS). Run this command after you log in as the machine administrator, open a new query in SSMS, and select the master database:

CREATE LOGIN [%COMPUTERNAME%\SQLRUserGroup] FROM WINDOWS

(Replace %COMPUTERNAME% with your VM name.)

To run SQL Server Management Studio, you can search for "SQL Server Management Studio" on the program list, or use Windows Search to find and run it. When prompted for credentials, select Windows Authentication, and use either the machine name or localhost in the SQL Server Name field.

How to use and run it

By default, the database server with the default database instance runs automatically. You can use tools like SQL Server Management Studio on the VM to access the SQL Server database locally. Local administrator accounts have admin access on the database.

Additionally, the DSVM comes with ODBC and JDBC drivers to talk to

  • SQL Server
  • Azure SQL databases
  • Azure Synapse Analytics resources from applications written in multiple languages, including Python and Machine Learning Server.

How is it configured and installed on the DSVM?

SQL Server is installed in the standard way. You can find it at C:\Program Files\Microsoft SQL Server. You can find the In-database Machine Learning Server instance at C:\Program Files\Microsoft SQL Server\MSSQL13.MSSQLSERVER\R_SERVICES. The DSVM also has a separate standalone Machine Learning Server instance, installed at C:\Program Files\Microsoft\R Server\R_SERVER. These two Machine Learning Server instances don't share libraries.

Ubuntu

You must first install SQL Server Developer Edition on an Ubuntu DSVM before you use it. Visit Quickstart: Install SQL Server and create a database on Ubuntu for more information.

Apache Spark 2.x (Standalone)

Category Value
What is it? A standalone (single node in-process) instance of the popular Apache Spark platform; a system for fast, large-scale data processing and machine-learning
Supported DSVM editions Linux
Typical uses
  • Rapid development of Spark/PySpark applications locally with a smaller dataset, and later deployment on large Spark clusters such as Azure HDInsight
  • Test Microsoft Machine Learning Server Spark context
  • Use SparkML or the Microsoft open-source MMLSpark library to build ML applications
Links to samples Jupyter sample:
  • ~/notebooks/SparkML/pySpark
  • ~/notebooks/MMLSpark

Microsoft Machine Learning Server (Spark context): /dsvm/samples/MRS/MRSSparkContextSample.R
Related tools on the DSVM
  • PySpark, Scala
  • Jupyter (Spark/PySpark Kernels)
  • Microsoft Machine Learning Server, SparkR, Sparklyr
  • Apache Drill

How to use it

You can run the spark-submit or pyspark command to submit Spark jobs on the command line. You can also create a new notebook with the Spark kernel to create a Jupyter notebook.

To use Spark from R, you use libraries like SparkR, Sparklyr, and Microsoft Machine Learning Server, which are available on the DSVM. See links to samples in the preceding table.

Setup

Before you run in a Spark context in Microsoft Machine Learning Server on Ubuntu Linux DSVM edition, you must complete a one-time setup step to enable a local single node Hadoop HDFS and Yarn instance. By default, Hadoop services are installed but disabled on the DSVM. To enable them, run these commands as root the first time:

echo -e 'y\n' | ssh-keygen -t rsa -P '' -f ~hadoop/.ssh/id_rsa
cat ~hadoop/.ssh/id_rsa.pub >> ~hadoop/.ssh/authorized_keys
chmod 0600 ~hadoop/.ssh/authorized_keys
chown hadoop:hadoop ~hadoop/.ssh/id_rsa
chown hadoop:hadoop ~hadoop/.ssh/id_rsa.pub
chown hadoop:hadoop ~hadoop/.ssh/authorized_keys
systemctl start hadoop-namenode hadoop-datanode hadoop-yarn

To stop the Hadoop-related services when you no longer need them, run systemctl stop hadoop-namenode hadoop-datanode hadoop-yarn.

A sample that demonstrates how to develop and test MRS in a remote Spark context (the standalone Spark instance on the DSVM) is provided and available in the /dsvm/samples/MRS directory.

How is it configured and installed on the DSVM?

Platform Install Location ($SPARK_HOME)
Linux /dsvm/tools/spark-X.X.X-bin-hadoopX.X

Libraries to access data from Azure Blob storage or Azure Data Lake Storage, using the Microsoft MMLSpark machine-learning libraries, are preinstalled in $SPARK_HOME/jars. These JARs are automatically loaded when Spark launches. By default, Spark uses data located on the local disk.

The Spark instance on the DSVM can access data stored in Blob storage or Azure Data Lake Storage. You must first create and configure the core-site.xml file, based on the template found in $SPARK_HOME/conf/core-site.xml.template. You must also have the appropriate credentials to access Blob storage and Azure Data Lake Storage. The template files use placeholders for Blob storage and Azure Data Lake Storage configurations.

For more information about creation of Azure Data Lake Storage service credentials, visit Authentication with Azure Data Lake Storage Gen1. After you enter the credentials for Blob storage or Azure Data Lake Storage in the core-site.xml file, you can reference the data stored in those sources through the URI prefix of wasb:// or adl://.