Sandbox Mode

See the result before you commit it

See the result before you commit it. No surprises. No blind runs.

ondatrasql sandbox

Mental Model

Production = read-only
Sandbox = writable copy
Your pipeline runs against both

You see the diff before anything is committed.

How It Works

Production catalog mounted read-only
Temporary sandbox catalog created
Models run normally — reads from production, writes to sandbox
Results compared
Sandbox discarded

What Makes It Different

This is not just a query preview. It’s a full simulation.

Runs the entire DAG — not just one model
Applies schema evolution
Executes validation (constraints + audits)
Shows downstream impact

This is a real run — just not committed.

Without Sandbox

Run pipeline
Something breaks
Debug in production

With Sandbox

Preview everything
See impact
Then commit

What You See

For each model:

Row changes (added / removed / percentage)
Schema changes (new columns, type changes)
Sample diffs
Validation results
Downstream effects

╠═══ Changes ══════════════════════════════════════════════════╣
║  staging.orders                                              ║
║    Rows: 100 → 142 (+42, +42.0%)                            ║
║                                                              ║
║  mart.revenue                                                ║
║    SCHEMA EVOLUTION: + Added: region (VARCHAR)               ║
║    Rows: 5 → 6 (+1, +20.0%)                                 ║

Failures Are Safe

If a model fails:

Nothing is committed
Downstream models are not affected
Production never sees it

You fix it before it’s real.

Full DAG Preview

The entire pipeline executes:

[OK] raw.customers (45ms)
[OK] staging.orders (95ms)
[OK] mart.revenue (54ms)
[FAIL] mart.kpis — constraint violated: total >= 0

You see exactly where things break.

Commands

ondatrasql sandbox                # All models
ondatrasql sandbox staging.orders # One model

Maintenance commands also support preview:

ondatrasql merge sandbox
ondatrasql expire sandbox

When to Use It

Before deploying changes
After modifying SQL
When adding new models
When debugging data issues

Or simply: whenever you want certainty.

Why This Matters

Most tools run pipelines blind. You find out something broke after it’s committed.

Sandbox lets you run your pipeline before you run your pipeline.