Migrating Dbt Project Across Platforms

Migrating a dbt Project Across Data Platforms

This skill guides migration of a dbt project from one data platform (source) to another (target) — for example, Snowflake to Databricks, or Databricks to Snowflake.

The core approach: dbt Fusion compiles SQL in real-time and produces rich, detailed error logs that tell you exactly what's wrong and where. We trust Fusion entirely for dialect conversion — no need to pre-document every SQL pattern difference. The workflow is: read Fusion's errors, fix them, recompile, repeat until done. Combined with dbt unit tests (generated on the source platform before migration), we prove both compilation correctness and data correctness on the target platform.

Success criteria: Migration is complete when:

1. dbtf compile finishes with 0 errors and 0 warnings on the target platform
2. All unit tests pass on the target platform (dbt test --select test_type:unit)
3. All models run successfully on the target platform (dbtf run)

Validation cost: Use dbtf compile as the primary iteration gate — it's free (no warehouse queries) and catches both errors and warnings from static analysis. Only dbtf run and dbt test incur warehouse cost; run those only after compile is clean.

Contents

• Additional Resources — Reference docs for installation, unit tests, profile targets
• Migration Workflow — 7-step migration process with progress checklist
• Don't Do These Things — Critical guardrails
• Known Limitations & Gotchas — Fusion-specific and cross-platform caveats

Additional Resources

• Installing dbt Fusion — How to install and verify dbt Fusion
• Generating Unit Tests — How to generate unit tests on the source platform before migration
• Switching Targets — How to configure the dbt target for the destination platform and update sources

Migration Workflow

Progress Checklist

Copy this checklist to track migration progress:

Migration Progress:
- [ ] Step 1: Verify dbt Fusion is installed and working
- [ ] Step 2: Assess source project (dbtf compile — 0 errors on source)
- [ ] Step 3: Generate unit tests on source platform
- [ ] Step 4: Switch dbt target to destination platform
- [ ] Step 5: Run Fusion compilation and fix all errors (dbtf compile — 0 errors on target)
- [ ] Step 6: Run and validate unit tests on target platform
- [ ] Step 7: Final validation and document changes in migration_changes.md

Instructions

When a user asks to migrate their dbt project to a different data platform, follow these steps. Create a migration_changes.md file documenting all code changes (see template below).

Step 1: Verify dbt Fusion is installed

Fusion is required — it provides the real-time compilation and rich error diagnostics that power this migration. Fusion may be available as dbtf or as dbt.

To detect which command to use:

1. Check if dbtf is available — if it exists, it's Fusion
2. If dbtf is not found, run dbt --version — if the output starts with dbt-fusion, then dbt is Fusion

Use whichever command is Fusion everywhere this skill references dbtf. If neither provides Fusion, guide the user through installation. See references/installing-dbt-fusion.md for details.

Step 2: Assess the source project

Run dbtf compile on the source platform target to confirm the project compiles cleanly with 0 errors. This establishes the baseline.

dbtf compile

If there are errors on the source platform, those must be resolved first before starting the migration. The migrating-dbt-core-to-v2 skill can help resolve v2 compatibility issues.

Step 3: Generate unit tests on source platform

While still connected to the source platform, generate dbt unit tests for key models to capture expected data outputs as a "golden dataset." These tests will prove data consistency after migration.

Which models to test: You must test every leaf node — models at the very end of the DAG that no other model depends on via ref(). Do not guess leaf nodes from naming conventions — derive them programmatically using the methods in references/generating-unit-tests.md. List all leaf nodes explicitly and confirm the count before writing tests. Also test any mid-DAG model with significant transformation logic (joins, calculations, case statements).

How to generate tests:

1. Identify leaf nodes: dbt ls --select "+tag:core" --resource-type model or inspect the DAG
2. Use dbt show --select model_name --limit 5 to preview output rows on the source platform
3. Pick 2-3 representative rows per model that exercise key business logic
4. Write unit tests in YAML using the dict format — see the adding-dbt-unit-test skill for detailed guidance on authoring unit tests
5. Place unit tests in the model's YAML file or a dedicated _unit_tests.yml file

See references/generating-unit-tests.md for detailed strategies on selecting test rows and handling complex models.

Verify tests pass on source: Run dbt test --select test_type:unit on the source platform to confirm all unit tests pass before proceeding.

Step 4: Switch dbt target to destination platform

Add a new target output for the destination platform within the existing profile in profiles.yml, then set it as the active target. Do not change the profile key in dbt_project.yml.

1. Add a new output entry in profiles.yml under the existing profile for the destination platform
2. Set the target: key in the profile to point to the new output
3. Update source definitions (_sources.yml) if the database/schema names differ on the destination platform
4. Remove or update any platform-specific configurations (e.g., +snowflake_warehouse, +file_format: delta)

See references/switching-targets.md for detailed guidance.

Step 5: Run Fusion compilation and fix errors

This is the core migration step. First, clear the target cache to avoid stale schema issues from the source platform, then run dbtf compile against the target platform — Fusion will flag every dialect incompatibility at once.

rm -rf target/
dbtf compile

How to work through errors:

1. Read the error output carefully — Fusion's error messages are rich and specific. They tell you the exact file, line number, and nature of the incompatibility.
2. Group similar errors — Many errors will be the same pattern (e.g., the same unsupported function used in multiple models). Fix the pattern once, then apply across all affected files.
3. Fix errors iteratively — Make fixes, recompile, check remaining errors. Summarize progress (e.g., "Fixed 12 errors, 5 remaining").
4. Common categories of errors:

• SQL function incompatibilities — Functions that exist on one platform but not another (e.g., GENERATOR on Snowflake vs. sequence on Databricks, nvl2 vs. CASE WHEN)
• Type mismatches — Data type names that differ between platforms (e.g., VARIANT on Snowflake vs. STRING on Databricks)
• Syntax differences — Platform-specific SQL syntax (e.g., FLATTEN on Snowflake vs. EXPLODE on Databricks)
• Unsupported config keys — Platform-specific dbt config like +snowflake_warehouse or +file_format: delta
• Macro/package incompatibilities — Packages that behave differently across platforms

Trust Fusion's errors: The error logs are the primary guide. Do not try to anticipate or pre-fix issues that Fusion hasn't flagged — this leads to unnecessary changes. Fix exactly what Fusion reports.