databricks-pybuilder-plugin 0.1.4

Pybuilder plugin providing tasks for assets deployment.

databricks-pybuilder-plugin

The plugin is considered to be used to deploy assets to a Databricks environment.

The plugin is activated with the following command in your build.by:

use_plugin('pypi:databricks_pybuilder_plugin')

It provides a set of tasks for uploading resources, workspaces, deploying jobs, or installing locally built whl dependencies into a databricks cluster.

Deployment to Databricks

Automated deployment is implemented on the PyBuilder tasks basis.

The list of available tasks:

1. export_workspace - exporting a workspace. A workspace is considered to be a folder in Databricks holding a notebook, or a set of notebooks.

The task uploads the src/main/scripts/ content into a Databricks workspace. It overrides files of the same names and leaves other files as is.

By default, a git branch name is used as a nested folder of a workspace for uploading the content into the Databricks workspace, if it's available. The default folder is used otherwise. Use the branch parameter to import a workspace in your own folder:

pyb export_workspace -P branch={custom_directory}

The final output path would be /team_folder/application_name/{custom_directory} in this way.

Executing the command from a master branch

pyb export_workspace

would upload the workspace files into /team_folder/application_name/master.

Here is the list of related deployment settings

Property	Value	Description
project_workspace_path	src/main/scripts/	The path to a folder in the project tree holding notebooks.
remote_workspace_path	/team_folder/application_name/	The Databricks folder the notebooks would be uploaded into from project_workspace_path.

All of the properties could be overridden with a -P parameter.

Usage example:

pyb export_workspace [-P env={env}] [-P branch={branch}]

Environment specific properties Disabled by default.

2. export_resources - exporting resources into dbfs. Uploads resource files into dbfs if any. Existing files are to be overridden.

Here is the list of related deployment settings

Property	Value	Description
project_resources_path	src/main/resources/	The path to the project resources.

All of the properties could be overridden with a -P parameter.

Usage example:

pyb export_resources [-P env={env}] [-P branch={branch}]

3. install_library - deploying an whl-archive to a Databricks cluster. Uploads an whl archive to a Volume, and re-attaches the library to a cluster by name. Re-installing a new library version triggers the cluster starting to uninstall old libraries versions and to install a new one. Repetitive installations of a library of the same version don't start the cluster and don't upload files to s3. The library is just re-attached to a cluster in this way. Installed libraries in other branches are not affected. Reinstalling a library in the same branch deleting all files under the same branch directory in AWS s3. If a Cloud Provider other than AWS is used, feel free to override the task using the Cloud specific API.

Here is the list of related deployment settings

Property	Value	Description
remote_cluster_name	Test_cluster_name	The name of a remote Databricks cluster the library to be installed to.
attachable_lib_path	/Volume/	The Volume path to a folder holding the whl archives (dependencies).
attachable_lib_s3_path	s3://{bucket_name}/volume/libs/	The Volume s3 path to a folder holding the whl archives (dependencies).

All of the properties could be overridden with a -P parameter.

Usage example:

pyb install_library

4. deploy_to_cluster - a full deployment to a cluster. Runs the export_resources, export_workspace, install_library in a row.

Usage example:

pyb deploy_to_cluster

5. deploy_job - deploying a job to the Databricks by name. Please, make sure that the job is created on the Databricks side.

Executes export_resources and export_workspace tasks preliminarily. Updates the existing job using a job definition file. The definition file supports the jinja2 template syntax. Please, check documentation for details: https://jinja.palletsprojects.com/en/2.11.x/

Here is the list of related deployment settings

Property	Value	Description
job_definition_path	src/main/databricks/databricks_job_settings.json	The project path to a job definition file.

All of the properties could be overridden with a -P parameter.

Usage example:

pyb deploy_job [-P env={env}] [-P branch={branch}]

5. post_init - initializing env and branch properties for further tasks. Executed automatically as a dependent task.

To Run a notebook with a custom dependency

1. Build the whl-archive with thepyb command.

2. Deploy all the assets using the command pyb deploy_to_cluster.

3. Get to the target folder in the Databricks workspace.

4. Attach the notebook to a cluster and run the script.

All properties list

Property	Default Value	Description
databricks_credentials	{ 'dev': {'host': '', 'token': ''} 'qa': {'host': '', 'token': ''} 'prod': {'host': '', 'token': ''} }	Please specify credentials in the dictionary format: host and token.
default_environment	dev	There are 3 supported environments: dev, qa and prod.
project_workspace_path	src/main/scripts/	The directory content is going to be uploaded into a databricks workspace. These are considered to be notebook scripts.
remote_workspace_path		The databricks workspace that files in project_workspace_path are copied to.
include_git_branch_into_output_workspace_path	True	The flag enables adding an extra directory with the branch name to the remote_workspace_path. Requires git to be installed.
enable_env_sensitive_workspace_properties	False	The flag enables environment properties chosen by the env_config_workspace_path.
env_config_workspace_path	environment-settings/{env}.py	The path to a property file to be chosen as a env properties. By default env included into a file name is used to pick properties.
env_config_name	env	The expected environment properties file name. The env_config_workspace_path will be copied to databricks workspace with name.
with_dbfs_resources	False	The flage enables uploading resource files from the project_resources_path directory to databricks hdfs dbfs_resources_path.
project_resources_path	src/main/resources/	The local directory path holding resource files to be copied (txt, csv etc).
dbfs_resources_path		The output hdfs directory on databricks environment holding resources.
attachable_lib_path		The output Volume directory on databricks environment holding a built dependency (whl-archive). It must start with the /Volume/ prefix and point to an existing Volume instance.
attachable_lib_s3_path		The output AWS s3 directory mounted as a Databricks Volume.
clean_attachable_lib	False	The Databricks Volume libraries folder will be cleaned from stored files.
attachable_lib_envs	['dev']	The list of environments that requires a dependency attached to a databricks cluster. The dependency is preliminary must be uploaded to the attachable_library_path.
cluster_init_timeout	5 * 60	The timeout of waiting a databricks cluster while it changes its state (initiating, restarting etc).
remote_cluster_name		The name of a databricks cluster that dependency is attached to.
job_definition_path	'src/main/databricks/job_settings.json'	The path to a dataricks job configuration in a json format - https://docs.databricks.com/dev-tools/api/2.0/jobs.html. It supports Jinja template syntax in order to setup env sensitive properties. It also supports multiple jobs definitions - use a json array for that. the list of properties available by default: env, branch, remote_workspace_path, remote_workspace_path
deploy_single_job		The name of a job to be deployed. If your databricks job config contains multiple definitions, you can deploy just one of these jobs specifying a name of the particular job.
extra_rendering_args		Custom properties to be populated in the job definition file. Use a dicionary as an argument. For example: {'app_name': name}.