External Hive Metastore

Every Azure Databricks deployment has a central Hive metastore accessible by all clusters to persist table metadata. Instead of using the Azure Databricks Hive metastore, you have the option to use your existing external Hive metastore instance.

This topic describes how to set up Azure Databricks clusters that connect to existing Hive metastores. We provide information about recommended metastore setup and cluster configuration requirements, followed by instructions for configuring clusters to connect to an external metastore. The following table summarizes which Hive metastore versions are supported in each version of Databricks Runtime.

Databricks Runtime Version Hive 0.13 - 1.2.1 Hive 2.1 Hive 2.2 Hive 2.3
5.x Yes Yes Yes Yes
4.x Yes Yes Yes Yes
3.x Yes Yes No No

Important

  • SQL Server does not work as the underlying metastore database for Hive 2.1 and 2.2.
  • You can use a Hive 1.2.0 or 1.2.1 metastore of an HDInsight cluster as an external metastore. See Use external metadata stores in Azure HDInsight.
  • To use Azure Database for MySQL as an external metastore, you must change the value of the lower_case_table_names property from 1 (the default) to 2 in the server-side database configuration. For details, see Identifier Case Sensitivity.

Hive metastore setup

Note

The examples in this document use Azure SQL database as the underlying metastore database.

The metastore client running inside a cluster connects to your underlying metastore database directly using JDBC.

To test network connectivity from a cluster to the metastore, you can run the following command inside a notebook:

%sh
nc -vz <DNS name> <port>

where

  • <DNS name> is the server name of the Azure SQL database.
  • <port> is the port of the database.

Cluster configurations

Typically, you must set two sets of configuration options to connect a cluster to an external metastore:

Spark-specific options

You must set spark.sql.hive.metastore.version to the version of your Hive metastore and spark.sql.hive.metastore.jars with the way that Spark retrieves JARs used by the metastore client:

  • If you are using Hive 0.13, you do not need to set spark.sql.hive.metastore.jars.

  • If you are using Hive 1.2.0 or 1.2.1, set spark.sql.hive.metastore.jars to builtin.

  • For all other versions, it’s best to download the correct version of the metastore JARs and then set spark.sql.hive.metastore.jars to point to the downloaded JARs. Databricks doesn’t recommend setting spark.sql.hive.metastore.jars to maven, since cluster creation can be slowed by repeated downloads or can even fail if Maven is temporarily unavailable. The recommended procedure is:

    1. Create a cluster with spark.sql.hive.metastore.jars set to maven and spark.sql.hive.metastore.version to match the version of your metastore.

    2. Once the cluster is running, search the driver log and find a line like the following:

      17/11/18 22:41:19 INFO IsolatedClientLoader: Downloaded metastore jars to /local_disk0/tmp/hive-v2_1-2037efa9-99ed-45be-9a04-83be817f1f0e
      

      The directory /local_disk0/tmp/hive-v2_1-2037efa9-99ed-45be-9a04-83be817f1f0e is the location of downloaded JARs in the driver node of the cluster.

    3. Use %sh cp -r /local_disk0/tmp/hive-v2_1-2037efa9-99ed-45be-9a04-83be817f1f0e /dbfs/hive_metastore_jar (replacing with your cluster’s info) to copy this directory to a directory in DBFS called hive_metastore_jar through the Fuse client in the driver node.

    4. Create an init script that copies /dbfs/hive_metastore_jar to the local filesystem of the node, making sure to make the init script sleep a few seconds before it accesses the DBFS Fuse client. This ensures that the client is ready.

    5. Set spark.sql.hive.metastore.jars to use this directory. If your init script copies /dbfs/hive_metastore_jar to /databricks/hive_metastore_jars/, you should set spark.sql.hive.metastore.jars to /databricks/hive_metastore_jars/*. The location must include the trailing /*.

    6. Restart the cluster.

Hive-specific options

This section describes options specific to Hive.

When you connect to an external metastore using local mode, you must set the following Hive configuration options:

# JDBC connect string for a JDBC metastore
javax.jdo.option.ConnectionURL <mssql-connection-string>

# Username to use against metastore database
javax.jdo.option.ConnectionUserName <mssql-username>

# Password to use against metastore database
javax.jdo.option.ConnectionPassword <mssql-password>

# Driver class name for a JDBC metastore
javax.jdo.option.ConnectionDriverName com.microsoft.sqlserver.jdbc.SQLServerDriver

where

  • <mssql-connection-string> is the JDBC connection string (which you can get in the Azure portal). Note that you do not need to include username and password in the connection string, because these will be set by javax.jdo.option.ConnectionUserName and javax.jdo.option.ConnectionDriverName.
  • <mssql-username> and <mssql-password> specify the username and password of your Azure SQL database account that has read/write access to the database.

Note

For production environments, we recommend that you set hive.metastore.schema.verification to true. This prevents Hive metastore client from implicitly modifying the metastore database schema when the metastore client version does not match the metastore database version. When enabling this setting for metastore client versions lower than Hive 1.2.0, make sure that the metastore client has the write permission to the metastore database (to prevent the issue described in HIVE-9749).

  • For Hive metastore 1.2.0 and higher, hive.metastore.schema.verification.record.version must be set to true in order to enable hive.metastore.schema.verification.
  • For Hive metastore 2.1.1 and higher, hive.metastore.schema.verification.record.version must be explicitly enabled as it is set to false by default.

Set up an external metastore using the web UI

To set up an external metastore using the Databricks web UI:

  1. Click the Clusters button on the sidebar.

  2. Click Create Cluster.

  3. Click Show advanced settings, and navigate to the Spark tab.

  4. Enter the following Spark configuration options:

    Set the following configurations under Spark Config.

    # Hive-specific configuration options.
    # spark.hadoop prefix is added to make sure these Hive specific options propagate to the metastore client.
    # JDBC connect string for a JDBC metastore
    spark.hadoop.javax.jdo.option.ConnectionURL <mssql-connection-string>
    
    # Username to use against metastore database
    spark.hadoop.javax.jdo.option.ConnectionUserName <mssql-username>
    
    # Password to use against metastore database
    spark.hadoop.javax.jdo.option.ConnectionPassword <mssql-password>
    
    # Driver class name for a JDBC metastore
    spark.hadoop.javax.jdo.option.ConnectionDriverName com.microsoft.sqlserver.jdbc.SQLServerDriver
    
    # Spark specific configuration options
    spark.sql.hive.metastore.version <hive-version>
    # Skip this one if <hive-version> is 0.13.x.
    spark.sql.hive.metastore.jars <hive-jar-source>
    
  5. Continue your cluster configuration, following the instructions in Cluster Configurations.

  6. Click Create Cluster to create the cluster.

Set up an external metastore using an init script

Cluster Node Initialization Scripts can be a convenient alternative when you want your clusters to connect to your existing Hive metastore without explicitly setting required configurations.

To set up an external metastore using an init script, open a notebook and execute the following snippet. This snippet adds an init script external-metastore.sh (this name is not mandatory) to /databricks/init/<cluster-name>/ in Databricks File System (DBFS). Alternatively, you can use the DBFS REST API’s put operation to create the init script. This init script writes required configuration options to a configuration file named 00-custom-spark.conf in a JSON-like format under /databricks/driver/conf/ inside every node of the cluster, whenever a cluster with the name specified as <cluster-name> starts. Note that Databricks provides default Spark configurations in the /databricks/driver/conf/spark-branch.conf file. Configuration files in the /databricks/driver/conf directory apply in reverse alphabetical order. If you want to change the name of the 00-custom-spark.conf file, make sure that it continues to apply before the spark-branch.conf file.

Note

If you want to set up all of your clusters automatically to access the external metastore, you can use a global init script. Simply change the location of the init script from /databricks/init/<cluster-name>/external-metastore.sh to /databricks/init/external-metastore.sh.

%scala

dbutils.fs.put(
    "/databricks/init/<cluster-name>/external-metastore.sh",
    """#!/bin/sh
      |# Loads environment variables to determine the correct JDBC driver to use.
      |source /etc/environment
      |# Quoting the label (i.e. EOF) with single quotes to disable variable interpolation.
      |cat << 'EOF' > /databricks/driver/conf/00-custom-spark.conf
      |[driver] {
      |    # Hive specific configuration options.
      |    # spark.hadoop prefix is added to make sure these Hive specific options will propagate to the metastore client.
      |    # JDBC connect string for a JDBC metastore
      |    "spark.hadoop.javax.jdo.option.ConnectionURL" = "<mssql-connection-string>"
      |
      |    # Username to use against metastore database
      |    "spark.hadoop.javax.jdo.option.ConnectionUserName" = "<mssql-username>"
      |
      |    # Password to use against metastore database
      |    "spark.hadoop.javax.jdo.option.ConnectionPassword" = "<mssql-password>"
      |
      |    # Driver class name for a JDBC metastore
      |    "spark.hadoop.javax.jdo.option.ConnectionDriverName" = "com.microsoft.sqlserver.jdbc.SQLServerDriver"
      |
      |    # Spark specific configuration options
      |    "spark.sql.hive.metastore.version" = "<hive-version>"
      |    # Skip this one if <hive-version> is 0.13.x.
      |    "spark.sql.hive.metastore.jars" = "<hive-jar-source>"
      |}
      |EOF
      |""".stripMargin,
    overwrite = true
)

Troubleshooting

Clusters do not start (due to incorrect init script settings)
If a global init script for setting up the external metastore causes cluster creation failure, you can convert the global init script to a cluster-scoped init script and then debug the init script using that cluster.
Error in SQL statement: InvocationTargetException

This exception usually appears in the following cases:

  • External metastore JDBC connection information misconfigured.

    If you see the following error message pattern in the full exception stack trace, then you probably hit this case:

    Caused by: javax.jdo.JDOFatalDataStoreException: Unable to open a test connection to the given database. JDBC url = [...]
    

    Verify the configured hostname, port, username, password, and JDBC driver class name. Also, make sure that the username has the right privilege to access the metastore database.

  • External metastore database not properly initialized.

    If you see the following error message pattern in the full exception stack trace, then you probably hit this case:

    Required table missing : "`DBS`" in Catalog "" Schema "". DataNucleus requires this table to perform its persistence operations. [...]
    

    Verify that you created the metastore database and put the correct database name in the JDBC connection string. Then, start a new cluster with the following two extra Spark configuration options:

    datanucleus.autoCreateSchema true
    datanucleus.fixedDatastore false
    

    In this way, the Hive client library will try to create and initialize tables in the metastore database automatically when it tries to access them but finds them absent.

Error in SQL statement: AnalysisException: Unable to instantiate org.apache.hadoop.hive.metastore.HiveMetastoreClient
If you find the error The specified datastore driver (driver name) was not found in the CLASSPATH in the full exception stacktrace, the cluster is configured to use an incorrect JDBC driver.
Setting datanucleus.autoCreateSchema to true doesn’t work as expected

By default, Databricks also sets datanucleus.fixedDatastore to true, which prevents any accidental structural changes to the metastore databases. Therefore, the Hive client library cannot create metastore tables even if you set datanucleus.autoCreateSchema to true. This strategy is, in general, safer for production environments since it prevents the metastore database to be accidentally upgraded.

If you do want to use datanucleus.autoCreateSchema to help initialize the metastore database, make sure you set datanucleus.fixedDatastore to false. Also, you may want to flip both flags after initializing the metastore database to provide better protection to your production environment.