adb-dbx-demo

This is a sample project for Databricks, generated via cookiecutter.

While using this project, you need Python 3.X and pip or conda for package management.

Local environment setup

Instantiate a local Python environment via a tool of your choice. This example is based on conda, but you can use any environment management tool:

conda create -n adb_dbx_demo python=3.9
conda activate adb_dbx_demo

If you don't have JDK installed on your local machine, install it (in this example we use conda-based installation):

conda install -c conda-forge openjdk=11.0.15

Install project locally (this will also install dev requirements):

pip install -e ".[local,test]"

Running unit tests

For unit testing, please use pytest:

pytest tests/unit --cov

Please check the directory tests/unit for more details on how to use unit tests. In the tests/unit/conftest.py you'll also find useful testing primitives, such as local Spark instance with Delta support, local MLflow and DBUtils fixture.

Running integration tests

There are two options for running integration tests:

On an all-purpose cluster via dbx execute
On a job cluster via dbx launch

For quicker startup of the job clusters we recommend using instance pools (AWS, Azure, GCP).

For an integration test on all-purpose cluster, use the following command:

dbx execute <workflow-name> --cluster-name=<name of all-purpose cluster>

To execute a task inside multitask job, use the following command:

dbx execute <workflow-name> \
    --cluster-name=<name of all-purpose cluster> \
    --job=<name of the job to test> \
    --task=<task-key-from-job-definition>

For a test on a job cluster, deploy the job assets and then launch a run from them:

dbx deploy <workflow-name> --assets-only
dbx launch <workflow-name>  --from-assets --trace

Interactive execution and development on Databricks clusters

dbx expects that cluster for interactive execution supports %pip and %conda magic commands.
Please configure your workflow (and tasks inside it) in conf/deployment.yml file.
To execute the code interactively, provide either --cluster-id or --cluster-name.

dbx execute <workflow-name> \
    --cluster-name="<some-cluster-name>"

Multiple users also can use the same cluster for development. Libraries will be isolated per each user execution context.

Working with notebooks and Repos

To start working with your notebooks from a Repos, do the following steps:

Add your git provider token to your user settings in Databricks
Add your repository to Repos. This could be done via UI, or via CLI command below:

databricks repos create --url <your repo URL> --provider <your-provider>

This command will create your personal repository under /Repos/<username>/adb_dbx_demo. 3. Use git_source in your job definition as described here

CI/CD pipeline settings

Please set the following secrets or environment variables for your CI provider:

DATABRICKS_HOST
DATABRICKS_TOKEN

Testing and releasing via CI pipeline

To trigger the CI pipeline, simply push your code to the repository. If CI provider is correctly set, it shall trigger the general testing pipeline
To trigger the release pipeline, get the current version from the adb_dbx_demo/__init__.py file and tag the current code version:

git tag -a v<your-project-version> -m "Release tag for version <your-project-version>"
git push origin --tags

mshtelma/adb-dbx-demo