feat(experimentation): add the Databricks warehouse provider

[Zaimwa9]

• I have read the Contributing Guide.
• I have added information to docs/ if required so people know about the feature.
• I have filled in the "Changes" section below.
• I have filled in the "How did you test this code" section below.

Changes

Second half of the Databricks provider, stacked on #8700 (the statement client). Adds Databricks as a bring-your-own warehouse for experiment results, behind the databricks_warehouse flag.

• DatabricksWarehouse: config and credentials validation, verify, event name and stats lookups, exposure buckets and results, over the client from feat(experimentation): add a Databricks statement client #8700.
• Config: host, workspace_id, region, warehouse_id, catalog, schema (default flagsmith_exp). Pasted URLs and HTTP paths are normalised, and the workspace ID can come from a pasted ?o= URL. Verify checks the workspace ID against the x-databricks-org-id response header.
• DatabricksDialect plugs into the shared ResultsQueryBuilder. Time buckets don't depend on the session timezone.
• Migration 0017 adds the warehouse type. The databricks_warehouse flag gates creating a Databricks connection, switching an existing one to it, and testing unsaved Databricks details.

How did you test this code?

Unit tests cover the provider, dialect, serializers and views at 100% diff coverage, including the expected SQL for every query.

Live QA against a Databricks trial workspace, using the same events as a ClickHouse baseline: 18/18 API calls returned identical results. A wrong workspace ID was also checked live.

Separately, the warehouse-delivery work ran a pipeline (ingestion → Kafka → Zerobus) into the trial workspace, then compute_experiment_exposures and compute_experiment_results on an earlier revision of this branch, without the two review fixes. The computed totals matched the generated data. The results API endpoint was not exercised over pipeline-delivered rows.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

3 Skipped Deployments

Project	Deployment	Actions	Updated
docs	Ignored	Preview	Oct 8, 2026 3:20pm UTC
flagsmith-frontend-preview	Ignored	Preview	Oct 8, 2026 3:20pm UTC
flagsmith-frontend-staging	Ignored	Preview	Oct 8, 2026 3:20pm UTC

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.86%. Comparing base (7b1f778) to head (e3de701).

Additional details and impacted files

@@                         Coverage Diff                          @@
##           feat/databricks-warehouse-provider    #8703    +/-   ##
====================================================================
  Coverage                               98.85%   98.86%            
====================================================================
  Files                                    1668     1669     +1     
  Lines                                   69013    69332   +319     
====================================================================
+ Hits                                    68223    68542   +319     
  Misses                                    790      790            

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Zaimwa9]

@themis-blindfold review

[themis-blindfold]

⚖️ Themis review: ✅ Ship it

TL;DR: The Databricks provider is correctly registered, feature-gated at the connection API, and follows the existing warehouse read, verification, caching, and error-reporting paths. The completed API unit suites and documentation checks passed; no blocking review issues found.

Area	Score
🎯 Correctness	5/5
🧪 Test coverage	5/5
📐 Code quality	5/5
🚀 Product impact	4/5

📝 Walkthrough

• Warehouse connection API - adds the Databricks type, migration, generated schema, and organisation-scoped feature gate for create, type changes, and unsaved connection tests.
• Databricks provider - validates and normalises configuration, verifies service-principal access, and reads event metadata, exposure buckets, and experiment aggregates.
• SQL dialect - supplies Databricks parameter binding, aggregation, array unnesting, and UTC-epoch bucket expressions for the shared query builders.
• Tests - cover configuration, credentials, verification errors, feature-gate behaviour, generated SQL, caching, and result conversion.

🧪 How to verify

1. Run cd api && pytest tests/unit/experimentation/warehouses/test_databricks.py.
2. Run cd api && pytest tests/unit/experimentation/test_views.py tests/unit/experimentation/test_serializers.py.
3. Enable experimentation_warehouse_connection and databricks_warehouse for a test organisation, then create and test a connection with valid service-principal credentials.
4. Repeat with the Databricks flag disabled and confirm create, type-switch, and unsaved connection tests return validation errors.
5. Create an experiment with occurrence, count, sum, and mean metrics and confirm the exposure and result refreshes match the warehouse rows.
   Automate: retain the existing provider and view unit suites in the API test matrix.

Product take: Solid capability expansion for customers running experimentation data in Databricks. The targeted flag makes the rollout appropriately controlled.

🧭 Assumptions & unverified claims

No unverified assumptions or claims.

Databricks joins the warehouse club without upsetting the query bouncer. · reviewed at e3de701