Apache Spark connector: SQL Server & Azure SQL

The Apache Spark connector for SQL Server and Azure SQL is a high-performance connector that enables you to use transactional data in big data analytics and persist results for ad hoc queries or reporting. The connector allows you to use any SQL database, on-premises or in the cloud, as an input data source or output data sink for Spark jobs.

This library contains the source code for the Apache Spark Connector for SQL Server and Azure SQL.

Apache Spark is a unified analytics engine for large-scale data processing.

There are two versions of the connector available through Maven, a 2.4.x compatible version and a 3.0.x compatible version. Both versions can be found here and can be imported using the coordinates below:

You can also build the connector from source or download the jar from the Release section in GitHub. For the latest information about the connector, see SQL Spark connector GitHub repository.

Supported Features

• Support for all Spark bindings (Scala, Python, R)
• Basic authentication and Active Directory (AD) Key Tab support
• Reordered dataframe write support
• Support for write to SQL Server Single instance and Data Pool in SQL Server Big Data Clusters
• Reliable connector support for Sql Server Single Instance

Supported Options

The Apache Spark Connector for SQL Server and Azure SQL supports the options defined here: SQL DataSource JDBC

In addition following options are supported

Other bulk copy options can be set as options on the dataframe and will be passed to bulkcopy APIs on write

Performance comparison

Apache Spark Connector for SQL Server and Azure SQL is up to 15x faster than generic JDBC connector for writing to SQL Server. Performance characteristics vary on type, volume of data, options used, and may show run to run variations. The following performance results are the time taken to overwrite a SQL table with 143.9M rows in a spark dataframe. The spark dataframe is constructed by reading store_sales HDFS table generated using spark TPCDS Benchmark. Time to read store_sales to dataframe is excluded. The results are averaged over three runs.

Config

• Spark config: num_executors = 20, executor_memory = '1664 m', executor_cores = 2
• Data Gen config: scale_factor=50, partitioned_tables=true
• Data file store_sales with nr of rows 143,997,590

Environment

• SQL Server Big Data Cluster CU5
• master + 6 nodes
• Each node gen 5 server, 512 GB Ram, 4 TB NVM per node, NIC 10 GB

Commonly Faced Issues

java.lang.NoClassDefFoundError: com/microsoft/aad/adal4j/AuthenticationException

This issue arises from using an older version of the mssql driver (which is now included in this connector) in your hadoop environment. If you are coming from using the previous Azure SQL Connector and have manually installed drivers onto that cluster for Microsoft Entra authentication compatibility, you will need to remove those drivers.

Steps to fix the issue:

1. If you are using a generic Hadoop environment, check and remove the mssql jar: rm $HADOOP_HOME/share/hadoop/yarn/lib/mssql-jdbc-6.2.1.jre7.jar. If you are using Databricks, add a global or cluster init script to remove old versions of the mssql driver from the /databricks/jars folder, or add this line to an existing script: rm /databricks/jars/*mssql*

2. Add the adal4j and mssql packages. For example, you can use Maven but any way should work.

   Caution

   Do not install the SQL spark connector this way.

3. Add the driver class to your connection configuration. For example:

   connectionProperties = {
     `Driver`: `com.microsoft.sqlserver.jdbc.SQLServerDriver`
   }`
   

For more information and explanation, see the resolution to https://github.com/microsoft/sql-spark-connector/issues/26.

Get Started

The Apache Spark Connector for SQL Server and Azure SQL is based on the Spark DataSourceV1 API and SQL Server Bulk API and uses the same interface as the built-in JDBC Spark-SQL connector. This integration allows you to easily integrate the connector and migrate your existing Spark jobs by simply updating the format parameter with com.microsoft.sqlserver.jdbc.spark.

To include the connector in your projects, download this repository and build the jar using SBT.

Write to a new SQL Table

When using mode overwrite if you do not use the option truncate on recreation of the table, indexes will be lost. , a columnstore table would now be a heap. If you want to maintain existing indexing please also specify option truncate with value true. For example, .option("truncate","true").

server_name = "jdbc:sqlserver://{SERVER_ADDR}"
database_name = "database_name"
url = server_name + ";" + "databaseName=" + database_name + ";"

table_name = "table_name"
username = "username"
password = "password123!#" # Please specify password here

try:
  df.write \
    .format("com.microsoft.sqlserver.jdbc.spark") \
    .mode("overwrite") \
    .option("url", url) \
    .option("dbtable", table_name) \
    .option("user", username) \
    .option("password", password) \
    .save()
except ValueError as error :
    print("Connector write failed", error)

Append to SQL Table

try:
  df.write \
    .format("com.microsoft.sqlserver.jdbc.spark") \
    .mode("append") \
    .option("url", url) \
    .option("dbtable", table_name) \
    .option("user", username) \
    .option("password", password) \
    .save()
except ValueError as error :
    print("Connector write failed", error)

Specify the isolation level

This connector by default uses READ_COMMITTED isolation level when performing the bulk insert into the database. If you wish to override the isolation level, use the mssqlIsolationLevel option as shown below.

    .option("mssqlIsolationLevel", "READ_UNCOMMITTED") \

Read from SQL Table

jdbcDF = spark.read \
        .format("com.microsoft.sqlserver.jdbc.spark") \
        .option("url", url) \
        .option("dbtable", table_name) \
        .option("user", username) \
        .option("password", password).load()

Microsoft Entra authentication

Python Example with Service Principal

context = adal.AuthenticationContext(authority)
token = context.acquire_token_with_client_credentials(resource_app_id_url, service_principal_id, service_principal_secret)
access_token = token["accessToken"]

jdbc_db = spark.read \
        .format("com.microsoft.sqlserver.jdbc.spark") \
        .option("url", url) \
        .option("dbtable", table_name) \
        .option("accessToken", access_token) \
        .option("encrypt", "true") \
        .option("hostNameInCertificate", "*.database.windows.net") \
        .load()

Python Example with Active Directory Password

jdbc_df = spark.read \
        .format("com.microsoft.sqlserver.jdbc.spark") \
        .option("url", url) \
        .option("dbtable", table_name) \
        .option("authentication", "ActiveDirectoryPassword") \
        .option("user", user_name) \
        .option("password", password) \
        .option("encrypt", "true") \
        .option("hostNameInCertificate", "*.database.windows.net") \
        .load()

A required dependency must be installed in order to authenticate using Active Directory.

The format of user when using ActiveDirectoryPassword should be the UPN format, for example [email protected].

For Scala, the _com.microsoft.aad.adal4j_ artifact will need to be installed.

For Python, the _adal_ library will need to be installed. This is available via pip.

Check the sample notebooks for examples.

Support

The Apache Spark Connector for Azure SQL and SQL Server is an open-source project. This connector does not come with any Microsoft support. For issues with or questions about the connector, create an Issue in this project repository. The connector community is active and monitoring submissions.

Next steps

Visit the SQL Spark connector GitHub repository.

For information about isolation levels, see SET TRANSACTION ISOLATION LEVEL (Transact-SQL).