Databricks Deploy Integration

Overview

Deploy Databricks workloads using Asset Bundles for environment management.

Prerequisites

• Databricks CLI v0.200+
• Asset Bundle project structure
• Workspace access for target environments

Instructions

Step 1: Project Structure

my-databricks-project/
├── databricks.yml              # Main bundle configuration
├── resources/
│   ├── jobs.yml                # Job definitions
│   ├── pipelines.yml           # DLT pipeline definitions
│   └── clusters.yml            # Cluster policies
├── src/
│   ├── notebooks/              # Databricks notebooks
│   │   ├── bronze/
│   │   ├── silver/
│   │   └── gold/
│   └── python/                 # Python modules
│       └── etl/
├── tests/
│   ├── unit/
│   └── integration/
├── fixtures/                   # Test data
└── conf/
    ├── dev.yml                 # Dev overrides
    ├── staging.yml             # Staging overrides
    └── prod.yml                # Production overrides

Step 2: Main Bundle Configuration

# databricks.yml
bundle:
  name: data-platform

variables:
  catalog:
    description: Unity Catalog name
    default: dev_catalog
  warehouse_id:
    description: SQL Warehouse ID
    default: ""

include:
  - resources/*.yml

workspace:
  host: ${DATABRICKS_HOST}

artifacts:
  etl_wheel:
    type: whl
    path: ./src/python
    build: poetry build

targets:
  dev:
    default: true
    mode: development
    variables:
      catalog: dev_catalog
    workspace:
      root_path: /Users/${workspace.current_user.userName}/.bundle/${bundle.name}/dev

  staging:
    mode: development
    variables:
      catalog: staging_catalog
    workspace:
      root_path: /Shared/.bundle/${bundle.name}/staging
    run_as:
      service_principal_name: staging-sp

  prod:
    mode: production
    variables:
      catalog: prod_catalog
      warehouse_id: "abc123def456"
    workspace:
      root_path: /Shared/.bundle/${bundle.name}/prod
    run_as:
      service_principal_name: prod-sp
    permissions:
      - level: CAN_VIEW
        group_name: data-consumers
      - level: CAN_MANAGE_RUN
        group_name: data-engineers
      - level: CAN_MANAGE
        service_principal_name: prod-sp

Step 3: Job Definitions

# resources/jobs.yml
resources:
  jobs:
    etl_pipeline:
      name: "${bundle.name}-etl-${bundle.target}"
      description: "Main ETL pipeline for ${var.catalog}"

      tags:
        environment: ${bundle.target}
        team: data-engineering
        managed_by: asset_bundles

      schedule:
        quartz_cron_expression: "0 0 6 * * ?"
        timezone_id: "America/New_York"
        pause_status: ${bundle.target == "dev" ? "PAUSED" : "UNPAUSED"}

      email_notifications:
        on_failure:
          - oncall@company.com
        no_alert_for_skipped_runs: true

      parameters:
        - name: catalog
          default: ${var.catalog}
        - name: run_date
          default: ""

      tasks:
        - task_key: bronze_ingest
          job_cluster_key: etl_cluster
          notebook_task:
            notebook_path: ../src/notebooks/bronze/ingest.py
            base_parameters:
              catalog: "{{job.parameters.catalog}}"
              run_date: "{{job.parameters.run_date}}"

        - task_key: silver_transform
          depends_on:
            - task_key: bronze_ingest
          job_cluster_key: etl_cluster
          notebook_task:
            notebook_path: ../src/notebooks/silver/transform.py

        - task_key: gold_aggregate
          depends_on:
            - task_key: silver_transform
          job_cluster_key: etl_cluster
          python_wheel_task:
            package_name: etl
            entry_point: gold_aggregate
          libraries:
            - whl: ../artifacts/etl_wheel/*.whl

        - task_key: data_quality
          depends_on:
            - task_key: gold_aggregate
          job_cluster_key: etl_cluster
          notebook_task:
            notebook_path: ../src/notebooks/quality/validate.py

      job_clusters:
        - job_cluster_key: etl_cluster
          new_cluster:
            spark_version: "14.3.x-scala2.12"
            node_type_id: ${bundle.target == "prod" ? "Standard_DS4_v2" : "Standard_DS3_v2"}
            num_workers: ${bundle.target == "prod" ? 4 : 1}
            autoscale:
              min_workers: ${bundle.target == "prod" ? 2 : 1}
              max_workers: ${bundle.target == "prod" ? 10 : 2}
            spark_conf:
              spark.databricks.delta.optimizeWrite.enabled: "true"
              spark.databricks.delta.autoCompact.enabled: "true"
            custom_tags:
              ResourceClass: ${bundle.target == "prod" ? "production" : "development"}

Step 4: Deployment Commands

# Validate bundle
databricks bundle validate
databricks bundle validate -t staging
databricks bundle validate -t prod

# Deploy to development
databricks bundle deploy -t dev

# Deploy to staging
databricks bundle deploy -t staging

# Deploy to production (with confirmation)
databricks bundle deploy -t prod

# Deploy specific resources
databricks bundle deploy -t staging --resource etl_pipeline

# Destroy resources (cleanup)
databricks bundle destroy -t dev --auto-approve

Step 5: Run Management

# Run job manually
databricks bundle run -t staging etl_pipeline

# Run with parameters
databricks bundle run -t staging etl_pipeline \
  --params '{"catalog": "test_catalog", "run_date": "2024-01-15"}'

# Check deployment status
databricks bundle summary -t prod

# View deployed resources
databricks bundle summary -t prod --output json | jq '.resources.jobs'

Step 6: Blue-Green Deployment

# scripts/blue_green_deploy.py
from databricks.sdk import WorkspaceClient
import time

def blue_green_deploy(
    w: WorkspaceClient,
    job_name: str,
    new_config: dict,
    rollback_on_failure: bool = True,
) -> dict:
    """
    Deploy job using blue-green strategy.

    1. Create new job version
    2. Run validation
    3. Switch traffic
    4. Remove old version (or rollback)
    """
    # Find existing job
    jobs = [j for j in w.jobs.list() if j.settings.name == job_name]
    old_job = jobs[0] if jobs else None

    # Create new job with suffix
    new_name = f"{job_name}-new"
    new_config["name"] = new_name
    new_job = w.jobs.create(**new_config)

    try:
        # Run validation job
        run = w.jobs.run_now(job_id=new_job.job_id)
        result = w.jobs.get_run(run.run_id).wait()

        if result.state.result_state != "SUCCESS":
            raise Exception(f"Validation failed: {result.state.state_message}")

        # Success - rename jobs
        if old_job:
            w.jobs.update(
                job_id=old_job.job_id,
                new_settings={"name": f"{job_name}-old"}
            )

        w.jobs.update(
            job_id=new_job.job_id,
            new_settings={"name": job_name}
        )

        # Cleanup old job
        if old_job:
            w.jobs.delete(job_id=old_job.job_id)

        return {"status": "SUCCESS", "job_id": new_job.job_id}

    except Exception as e:
        if rollback_on_failure:
            # Cleanup new job
            w.jobs.delete(job_id=new_job.job_id)
        raise

Output

• Deployed Asset Bundle
• Jobs created in target workspace
• Environment-specific configurations applied

Error Handling

Examples

Environment Comparison

# Compare configurations across environments
databricks bundle summary -t dev --output json > dev.json
databricks bundle summary -t prod --output json > prod.json
diff <(jq -S . dev.json) <(jq -S . prod.json)

Rollback Procedure

# Quick rollback using git
git checkout HEAD~1 -- databricks.yml resources/
databricks bundle deploy -t prod --force

Resources

• Asset Bundles Documentation
• Bundle YAML Reference
• Deployment Best Practices

Next Steps

For webhooks and events, see databricks-webhooks-events.