Databricks on Azure

Destination Azure Databricks #

On this page, you’ll find step-by-step instructions on how to set up your Azure Databricks instance as data target with Arcion. The extracted replicant-cli will be referred to as the $REPLICANT_HOME directory in the following steps.

Prerequisites #

  • A Databricks account on Azure
  • Azure container in ADLS Gen2 (Azure Data Lake Storage Gen2)

After making sure of the prerequisities in the preceeding section, follow these steps to set up your Azure Databricks with Arcion.

I. Create a Databricks cluster #

To create a Databricks cluster, follow these steps:

  1. Log in to your Databricks account.

  2. In the Databricks console, click Compute.

  3. Click Create Cluster.

  4. Enter a cluster name of your choice.

  5. Select the latest Databricks runtime version.

  6. Click Create Cluster.

II. Get connection details for Databricks cluster #

To establish connection between your Databricks instance and Arcion, you need to provide the connection details for your cluster. The connection details are available from Databricks JDBC and ODBC drivers configuration page. To get the connection details, follow these steps:

  1. Navigate to Advanced Options and click the JDBC/ODBC tab.

  2. Make a note of the following values. These are necessary to configure Arcion Replicant for replication.

    • Server Hostname
    • Port
    • JDBC URL

III. Create a personal access token for the Databricks cluster #

To create a personal access token, see Generate a personal access token in Databricks documentation. Make a note of the token as it’s required to configure Arcion Replicant for replication.

IV. Configure ADLS container as stage #

  1. Frist you need to set up access to ADLS. You can set up access in the following two ways:

  2. Open your Databricks console and go to the cluster configuration page.

  3. Click Compute.

  4. Expand the Advanced Options section.

  5. In the Spark Config box, paste the following settings:

    spark.hadoop.fs.azure.account.auth.type.STORAGE_ACCOUNT_NAME.dfs.core.windows.net OAuth
spark.hadoop.fs.azure.account.oauth.provider.type.STORAGE_ACCOUNT_NAME.dfs.core.windows.net org.apache.hadoop.fs.azurebfs.oauth2.ClientCredsTokenProvider
spark.hadoop.fs.azure.account.oauth2.client.id.STORAGE_ACCOUNT_NAME.dfs.core.windows.net APPLICATION-ID
spark.hadoop.fs.azure.account.oauth2.client.secret.STORAGE_ACCOUNT_NAME.dfs.core.windows.net {{secrets/SECRET_SCOPE/KEY_STORED_IN_SECRET_SCOPE}}
spark.hadoop.fs.azure.account.oauth2.client.endpoint.STORAGE_ACCOUNT_NAME.dfs.core.windows.net https://login.microsoftonline.com/DIRECTORY_ID/oauth2/token

    Replace the following:

    • STORAGE_ACCOUNT_NAME: the name of your ADLS Gen2 storage account
    • APPLICATION_ID: the Application (client) ID for the Azure Active Directory (Azure AD) application
    • SECRET_SCOPE: the name of your client secret scope
    • KEY_STORED_IN_SECRET_SCOPE: the name of the key containing the client secret
    • DIRECTORY_ID: the Directory (tenant) ID for the Azure Active Directory (Azure AD) application

  6. Copy the SECRET_KEY from Azure portal. These keys are required for establishing a connection from Arcion Replicant to ADLS. Replicant only uses these credentials to upload files to or delete from ADLS container.

V. Obtain the JDBC Driver for Databricks #

Replicant requires the Databricks JDBC Driver as a dependency. To obtain the appropriate driver, follow the steps below:

  • Go to the Databricks JDBC Driver download page and download the driver.
  • From the downloaded ZIP, locate and extract the DatabricksJDBC42.jar file.
  • Put the DatabricksJDBC42.jar file inside $REPLICANT_HOME/lib directory.

VI. Configure Replicant connection for Databricks #

In this step, you need to provide the Databricks connection details to Arcion. To do so, follow these steps:

  1. You can find a sample connection configuration file databricks.yaml in the $REPLICANT_HOME/conf/conn/ directory.

  2. The connection configuration file has the following two parts:

    • Parameters related to target Databricks server connection.
    • Parameters related to stage configuration.

    If you store your Databricks server connection credentials in AWS Secrets Manager, you can tell Replicant to retrieve them. For more information, see Retrieve credentials from AWS Secrets Manager. Otherwise, you can put your credentials in plain form like the sample below:

    type: DATABRICKS_DELTALAKE

url: "JDBC_URL"
username: USERNAME
password: "PASSWORD"
host: "HOSTNAME"
port: "PORT_NUMBER"
max-connections: 30
max-metadata-connections: 10

    Replace the following:

    Feel free to change the values of max-connections and max-metadata-connections as you need.

    For Databricks Unity Catalog, set the connection type to DATABRICKS_LAKEHOUSE. For more information, see Databricks Unity Catalog Support.

    It is mandatory to use an external stage to hold the data files and load them on the target database from there. The stage section allows specifying the details Replicant needs to connect to and use a given stage.

    • type[v21.06.14.1]: The stage type. For Azure Databricks, the type is AZURE.
      For Databricks Unity Catalog, set type to DATABRICKS_LAKEHOUSE. For more information, see Databricks Unity Catalog Support.
    • root-dir: The directory created under ADLS container. This directory is used to stage bulk-load files.
    • conn-url[v21.06.14.1]: The name of the ADLS container.
    • account-name[v21.06.14.1]: The name of the ADLS storage account.
    • secret-key[v21.06.14.1]: The SECRET_KEY for the user with write/delete access on ADLS container. This is the last step when you configure ADLS container as stage.

    The following is a sample stage configuration for Azure Databricks:

    stage:
  type: AZURE
  root-dir: "replicate-stage/databricks-stage"
  conn-url: "replicant-container"
  account-name: "replicant-storageaccount"
  secret-key: "YOUR_SECRET_KEY"

VII. Set up Applier Configuration #

  1. From $REPLICANT_HOME, navigate to the applier configuration file:

    vi conf/dst/databricks.yaml

  2. The configuration file has two parts:

    • Parameters related to snapshot mode.
    • Parameters related to realtime mode.

    For snapshot mode, make the necessary changes as follows:

    snapshot:
  threads: 16 #Maximum number of threads Replicant should use for writing to the target

  #If bulk-load is used, Replicant will use the native bulk-loading capabilities of the target database
  bulk-load:
    enable: true
    type: FILE
    serialize: true|false #Set to true if you want the generated files to be applied in serial/parallel fashion

    There are some additional parameters available that you can use in snapshot mode:

    snapshot:
  enable-optimize-write: true
  enable-auto-compact:  true
  enable-unmanaged-delta-table: false
  unmanaged-delta-table-location:
  init-sk: false
  per-table-config:
    init-sk: false
    shard-key:
    enable-optimize-write: true
    enable-auto-compact: true
    enable-unmanaged-delta-table: false
    unmanaged-delta-table-location:

    These parameters are specific to Databricks as destination. More details about these parameters are as follows:

    • enable-optimize-write: Databricks dynamically optimizes Apache Spark partition sizes based on the actual data, and attempts to write out 128 MB files for each table partition. This is an approximate size and can vary depending on dataset characteristics.

      Default: By default, this parameter is set to true.

    • enable-auto-compact: After an individual write, Databricks checks if files can be compacted further. If so, it runs an OPTIMIZE job to further compact files for partitions that have the most number of small files. The job is run with 128 MB file sizes instead of the 1 GB file size used in the standard OPTIMIZE.

      Default: By default, this parameter is set to true.

    • enable-unmanaged-delta-table: An unmanaged table is a Spark SQL table for which Spark manages only the metadata. The data is stored in the path provided by the user. So when you perform DROP TABLE <example-table>, Spark removes only the metadata and not the data itself. The data is still present in the path you provided.

      Default: By default, this parameter is set to false.

    • unmanaged-delta-table-location: The path where data for the unmanaged table is to be stored. It can be a Databricks DBFS path (for example FileStore/tables), or an S3 path (for example, s3://replicate-stage/unmanaged-table-data) where the S3 bucket is accessible to Databricks.

    • init-sk: Partition-key on the source table is represented as a shard-key by replicant. By default the target table does not include this sharding information. If init-sk is true we add the shard-key/partition key to target table create SQL. Shard-key replication is disabled by default because DML replication with partitioned tables in Databricks is very slow if the partition key has a high distinct count.

      Default: By default, this parameter is set to false.

    • per-table-config: This configuration allows you to specify various properties for target tables on a per table basis.

      • init-sk: Partition-key on the source table is represented as a shard-key by replicant. By default, the target table does not include this sharding information. If init-sk is true we add the shard-key/partition key to target table create SQL. Shard-key replication is disabled by default because DML replication with partitioned tables in\ databricks is very slow if the partition key has a high distinct count.

        Default: By default, this parameter is set to false.

      • shard-key: Shard key to be used for partitioning the target table.

      • enable-optimize-write: Databricks dynamically optimizes Apache Spark partition sizes based on the actual data, and attempts to write out 128 MB files for each table partition. This is an approximate size and can vary depending on dataset characteristics.

        Default: By default, this parameter is set to true.

      • enable-auto-compact: After an individual write, Databricks checks if files can be compacted further. If so, it runs an OPTIMIZE job to further compact files for partitions that have the most number of small files. The job is run with 128 MB file sizes instead of the 1 GB file size used in the standard OPTIMIZE.

        Default: By default, this parameter is set to true.

      • enable-unmanaged-delta-table: An unmanaged table is a Spark SQL table for which Spark manages only the metadata. The data is stored in the path provided by the user. So when you perform DROP TABLE <example-table>, Spark removes only the metadata and not the data itself. The data is still present in the path you provided.

        Default: By default, this parameter is set to false.

      • unmanaged-delta-table-location: The path where data for the unmanaged table is to be stored. It can be a Databricks DBFS path (for example FileStore/tables), or an S3 path (for example, s3://replicate-stage/unmanaged-table-data) where the S3 bucket is accessible to Databricks.

    If you want to operate in realtime mode, you can use the realtime section to specify your configuration. For example:

    realtime:
  threads: 4 #Maximum number of threads Replicant should use for writing to the target

    Enabling Type-2 CDC #

    From version 22.07.19.3 onwards, Arcion supports Type-2 CDC for Databricks as the Target. Type-2 CDC enables a Target to have a history of all transactions performed in the Source. For example:

    • An INSERT in the Source is an INSERT in the Target.
    • An UPDATE in the Source is an INSERT in the Target with additional metadata like Operation Performed, Time of Operation, etc.
    • A DELETE in the Source is an INSERT in the Target: INSERT with OPER_TYPE as DELETE.

    Arcion supports the following metadata related to source-specific fields:

    • query_timestamp: Time at which the user on Source fired a query.
    • extraction_timestamp: Time at which Replicant detected the DML from logs.
    • OPER_TYPE: Type of the operation (INSERT/UPDATE/DELETE).

    The primary requirement for Type-2 CDC is to enable full row logging in the Source.

    Support for Type-2 CDC is limited to the following cases:

    • Sources that support CDC.
    • realtime and full modes.

    To enable Type-2 CDC for your Databricks target, follow the steps below:

    1. Add the following two parameters under the realtime section of the Databricks Applier configuration file:

      realtime:
  enable-type2-cdc: true
  replay-strategy: NONE

    2. In the Extractor configuration file of Source, add the following parameter under the snapshot section:

      snapshot:
  csv-publish-method: READ

For a detailed explanation of configuration parameters in the Applier file, read Applier Reference.

Databricks Unity Catalog Support (Beta) #

Note: This feature is in beta.

From version 22.08.31.3 onwards, Arcion has added support for Databricks Unity Catalog. The support is still in beta phase, with complete support to land gradually in future releases.

As of now, note the following about the state of Arcion’s Unity Catalog support:

  • Legacy Databricks only supports two-level namespace:

    • Schemas
    • Tables

    With introduction of Unity Catalog, Databricks now exposes a three-level namespace that organizes data.

    • Catalogs
    • Schemas
    • Tables

    Arcion adds support for Unity Catalog by introducing a new child storage type (DATABRICKS_LAKEHOUSE child of DATABRICKS_DELTALAKE).

  • If you’re using Unity Catalog, notice the following when configuring your Target Databricks with Arcion:

    • Set the connection type to DATABRICKS_LAKEHOUSE in the connection configuration file.
    • To avoid manual steps to configure staging, Databricks has introduced personal staging. To read the staging URL, we’ve added a new configuration parameter UNITY_CATALOG_PERSONAL_STAGE. The complete stage configuration is as follows: 
      stage:
  type: UNITY_CATALOG_PERSONAL_STAGE
  staging-url: STAGING_URL
  file-format: DATA_FILE_FORMAT
      Replace the following:

      • STAGING_URL: the temporary staging URL—for example, stage://tmp/userName/rootDir.

      • DATA_FILE_FORMAT: the type of data file format. Supported formats are PARQUET and CSV.

        Default: PARQUET.

  • We use SparkJDBC42 driver for Legacy Databricks (DATABRICKS_DELTALAKE) and DatabricksJDBC42 for Unity catalog (DATABRICKS_LAKEHOUSE). For instructions on how to obtain these drivers, see Obtain the JDBC Driver for Databricks.

  • Replicant supports Unity Catalog on AWS and Azure platforms.