JosiahSiegel/databricks-2025

Azure Data Factory Databricks Integration 2025

Databricks Job Activity (Recommended 2025)

CRITICAL UPDATE (2025): The Databricks Job activity is now the ONLY recommended method for orchestrating Databricks in ADF. Microsoft strongly recommends migrating from legacy Notebook, Python, and JAR activities.

Quick Reference

• Activity type: DatabricksJob (NOT DatabricksSparkJob or DatabricksNotebook)
• Parameter property: jobParameters (NOT parameters)
• Linked service auth: Managed Identity ("authentication": "MSI") recommended
• Cluster config: Do NOT specify cluster properties in linked service; the Databricks Job controls compute

Why Databricks Job Activity?

| Feature | Notebook Activity (Legacy) | Job Activity (2025) | |---------|---------------------------|---------------------| | Compute | Must configure cluster in linked service | Serverless by default | | Workflow tasks | Single notebook | Multi-task DAGs (notebook, Python, SQL, DLT) | | Retry | ADF-level only | Job-level + task-level | | Repair runs | Not supported | Rerun failed tasks only | | Git integration | Limited | Full Databricks Git support + DABs | | Lineage | None | Built-in data lineage | | If/Else logic | Must use ADF control flow | Native If/Else task types |

Benefits Summary

1. Serverless Execution -- No cluster specification needed; automatic serverless compute with faster startup and lower costs
2. Advanced Workflow Features -- Run As, Task Values, Conditional Execution, AI/BI Tasks, Repair Runs, Notifications, Queuing
3. Centralized Job Management -- Jobs defined once in Databricks workspace; single source of truth with Git-backed versioning
4. Cost Optimization -- Serverless compute (pay only for execution), job clusters (auto-terminating), spot instance support

For complete JSON examples of Job activity, linked service, and pipeline configurations, see references/databricks-job-examples.md.

Connectors and Enhancements (2025+)

ServiceNow V2 Connector (RECOMMENDED - V1 End of Support)

ServiceNow V1 connector is at End of Support. Migrate to V2 immediately.

| Feature | V1 | V2 | |---------|----|----| | Linked service type | ServiceNow | ServiceNowV2 | | Source type | ServiceNowSource | ServiceNowV2Source | | Query builder | Custom | Aligns with ServiceNow condition builder | | Performance | Standard | Enhanced extraction | | OData support | No | Yes |

Migration steps: Update linked service type to ServiceNowV2, update source type to ServiceNowV2Source, test queries in ServiceNow UI condition builder, adjust timeouts.

Enhanced PostgreSQL Connector

Improved performance with 2025 SSL enhancements: enableSsl: true, sslMode: "Require".

Enhanced Snowflake Connector

Improved performance with KeyPair authentication support and Key Vault secret integration.

Managed Identity for Azure Storage

New managed identity support for Azure Table Storage and Azure Files connectors (system-assigned and user-assigned).

Mapping Data Flows - Spark 3.3

Spark 3.3 now powers Mapping Data Flows with 30% faster processing, Adaptive Query Execution (AQE), dynamic partition pruning, improved caching, and better column statistics.

Azure DevOps Server 2022 Support

Git integration now supports on-premises Azure DevOps Server 2022 via the hostName property.

For complete JSON examples of all connectors, see references/connector-examples.md.

Managed Identity 2025 Best Practices

User-Assigned vs System-Assigned

| Scenario | Recommendation | |----------|---------------| | Single ADF, simple setup | System-assigned | | Multiple data factories | User-assigned (shared identity) | | Complex multi-environment | User-assigned | | Granular permission control | User-assigned | | Identity lifecycle independence | User-assigned |

Use ADF's centralized Credentials feature to consolidate Microsoft Entra ID-based credentials across multiple linked services.

MFA Enforcement (Enforced Since October 2025)

Azure MFA is mandatory for all interactive user logins. Impact on ADF:

• Managed identities are UNAFFECTED -- no MFA required for service accounts
• Service principals with certificate auth are the recommended alternative to secrets
• All interactive user logins require MFA

Principle of Least Privilege

| Resource | Source Role | Sink Role | |----------|-----------|-----------| | Storage Blob | Storage Blob Data Reader | Storage Blob Data Contributor | | SQL Database | db_datareader | db_datareader + db_datawriter | | Key Vault | Get secrets only | Get secrets only |

For complete managed identity JSON examples, see references/connector-examples.md.

Best Practices (2025)

1. Use Databricks Job Activity (MANDATORY) -- Stop using Notebook, Python, JAR activities. Define workflows in Databricks workspace with serverless compute.

2. Managed Identity Authentication (MANDATORY) -- Use managed identities for ALL Azure resources. Leverage Credentials feature for consolidation. MFA-compliant since October 2025.

3. Monitor Job Execution -- Track Databricks Job run IDs from ADF output, log parameters for auditability, set up alerts for failures, leverage built-in lineage.

4. Optimize Spark 3.3 (Data Flows) -- Enable AQE, use 4-8 partitions per core, broadcast joins for small dimensions, dynamic partition pruning.

Resources

• Databricks Job Activity
• ADF Connectors
• Managed Identity Authentication
• Mapping Data Flows

Progressive Disclosure References

• Databricks Job Examples: references/databricks-job-examples.md - Complete JSON for Job activity, linked services, pipeline, and Databricks workspace job definition
• Connector Examples: references/connector-examples.md - Complete JSON for ServiceNow V2, PostgreSQL, Snowflake, Azure Storage MI, Mapping Data Flows, and Azure DevOps Server