Skip to main content

Apache Spark setup

profiles.yml file is for dbt Core users only

If you're using dbt Cloud, you don't need to create a profiles.yml file. This file is only for dbt Core users. To connect your data platform to dbt Cloud, refer to About data platforms.

If you're using Databricks, use dbt-databricks

If you're using Databricks, the dbt-databricks adapter is recommended over dbt-spark. If you're still using dbt-spark with Databricks consider migrating from the dbt-spark adapter to the dbt-databricks adapter.

For the Databricks version of this page, refer to Databricks setup.

  • Maintained by: dbt Labs
  • Authors: core dbt maintainers
  • GitHub repo: dbt-labs/dbt-spark
  • PyPI package: dbt-spark
  • Slack channel: db-databricks-and-spark
  • Supported dbt Core version: v0.15.0 and newer
  • dbt Cloud support: Supported
  • Minimum data platform version: n/a

Installing dbt-spark

Use pip to install the adapter. Before 1.8, installing the adapter would automatically install dbt-core and any additional dependencies. Beginning in 1.8, installing an adapter does not automatically install dbt-core. This is because adapters and dbt Core versions have been decoupled from each other so we no longer want to overwrite existing dbt-core installations. Use the following command for installation:

Configuring dbt-spark

For Spark-specific configuration, please refer to Spark configs.

If connecting to Databricks via ODBC driver, it requires pyodbc. Depending on your system, you can install it seperately or via pip. See the pyodbc wiki for OS-specific installation details.

If connecting to a Spark cluster via the generic thrift or http methods, it requires PyHive.

# odbc connections
$ python -m pip install "dbt-spark[ODBC]"

# thrift or http connections
$ python -m pip install "dbt-spark[PyHive]"
# session connections
$ python -m pip install "dbt-spark[session]"

Configuring dbt-spark

For Spark-specific configuration please refer to Spark Configuration

For further info, refer to the GitHub repository: dbt-labs/dbt-spark

Connection methods​

dbt-spark can connect to Spark clusters by four different methods:

  • odbc is the preferred method when connecting to Databricks. It supports connecting to a SQL Endpoint or an all-purpose interactive cluster.

  • thrift connects directly to the lead node of a cluster, either locally hosted / on premise or in the cloud (e.g. Amazon EMR).

  • http is a more generic method for connecting to a managed service that provides an HTTP endpoint. Currently, this includes connections to a Databricks interactive cluster.

  • session connects to a pySpark session, running locally or on a remote machine.

Advanced functionality

The session connection method is intended for advanced users and experimental dbt development. This connection method is not supported by dbt Cloud.

ODBC​

Use the odbc connection method if you are connecting to a Databricks SQL endpoint or interactive cluster via ODBC driver. (Download the latest version of the official driver here.)

~/.dbt/profiles.yml
your_profile_name:
  target: dev
  outputs:
    dev:
      type: spark
      method: odbc
      driver: [path/to/driver]
      schema: [database/schema name]
      host: [yourorg.sparkhost.com]
      organization: [org id]    # Azure Databricks only
      token: [abc123]
      
      # one of:
      endpoint: [endpoint id]
      cluster: [cluster id]
      
      # optional
      port: [port]              # default 443
      user: [user]
      server_side_parameters:
        "spark.driver.memory": "4g"

Thrift​

Use the thrift connection method if you are connecting to a Thrift server sitting in front of a Spark cluster, e.g. a cluster running locally or on Amazon EMR.

~/.dbt/profiles.yml
your_profile_name:
  target: dev
  outputs:
    dev:
      type: spark
      method: thrift
      schema: [database/schema name]
      host: [hostname]
      
      # optional
      port: [port]              # default 10001
      user: [user]
      auth: [e.g. KERBEROS]
      kerberos_service_name: [e.g. hive]
      use_ssl: [true|false]   # value of hive.server2.use.SSL, default false
      server_side_parameters:
        "spark.driver.memory": "4g"

HTTP​

Use the http method if your Spark provider supports generic connections over HTTP (e.g. Databricks interactive cluster).

~/.dbt/profiles.yml
your_profile_name:
  target: dev
  outputs:
    dev:
      type: spark
      method: http
      schema: [database/schema name]
      host: [yourorg.sparkhost.com]
      organization: [org id]    # Azure Databricks only
      token: [abc123]
      cluster: [cluster id]
      
      # optional
      port: [port]              # default: 443
      user: [user]
      connect_timeout: 60       # default 10
      connect_retries: 5        # default 0
      server_side_parameters:
        "spark.driver.memory": "4g"

Databricks interactive clusters can take several minutes to start up. You may include the optional profile configs connect_timeout and connect_retries, and dbt will periodically retry the connection.

Session​

Use the session method if you want to run dbt against a pySpark session.

~/.dbt/profiles.yml
your_profile_name:
  target: dev
  outputs:
    dev:
      type: spark
      method: session
      schema: [database/schema name]
      host: NA                           # not used, but required by `dbt-core`
      server_side_parameters:
        "spark.driver.memory": "4g"

Optional configurations​

Retries​

Intermittent errors can crop up unexpectedly while running queries against Apache Spark. If retry_all is enabled, dbt-spark will naively retry any query that fails, based on the configuration supplied by connect_timeout and connect_retries. It does not attempt to determine if the query failure was transient or likely to succeed on retry. This configuration is recommended in production environments, where queries ought to be succeeding.

For instance, this will instruct dbt to retry all failed queries up to 3 times, with a 5 second delay between each retry:

~/.dbt/profiles.yml
retry_all: true
connect_timeout: 5
connect_retries: 3

Caveats​

Usage with EMR​

To connect to Apache Spark running on an Amazon EMR cluster, you will need to run sudo /usr/lib/spark/sbin/start-thriftserver.sh on the master node of the cluster to start the Thrift server (see the docs for more information). You will also need to connect to port 10001, which will connect to the Spark backend Thrift server; port 10000 will instead connect to a Hive backend, which will not work correctly with dbt.

Supported functionality​

Most dbt Core functionality is supported, but some features are only available on Delta Lake (Databricks).

Delta-only features:

  1. Incremental model updates by unique_key instead of partition_by (see merge strategy)
  2. Snapshots
  3. Persisting column-level descriptions as database comments

Default namespace with Thrift connection method​

If your Spark cluster doesn't have a default namespace, metadata queries that run before any dbt workflow will fail, causing the entire workflow to fail, even if your configurations are correct. The metadata queries fail there's no default namespace in which to run it.

To debug, review the debug-level logs to confirm the query dbt is running when it encounters the error: dbt run --debug or logs/dbt.log.

0
Edit this page