Migrations with Alembic
Autogenerate, review, and apply schema revisions.
Recipe
Quick-reference recipe card - copy-paste ready.
alembic revision --autogenerate -m "add column"
alembic upgrade headWhen to reach for this:
- Every schema change tracked
- Multi-environment deploys
- Rollback planning
Working Example
def upgrade():
op.add_column("items", sa.Column("sku", sa.String(32), nullable=True))What this demonstrates:
- autogenerate workflow
- upgrade operations
- revision chain
Deep Dive
How It Works
- Alembic compares metadata to live DB.
- Human-review autogen before merge.
- Keep migrations backward compatible when zero-downtime.
Gotchas
- Boundary validation skipped - Invalid data reaches persistence layers.. Fix: Validate with Pydantic or framework forms at the edge..
- Leaking stack traces - Clients see internal errors.. Fix: Map exceptions to stable HTTP responses..
- Blocking async event loops - Workers stall under concurrent load.. Fix: Use async drivers or threadpool wrappers..
- Secrets in source control - Credentials leak via git history.. Fix: Load secrets from env or a vault at runtime..
- Missing observability - Incidents are hard to debug.. Fix: Add structured logs, metrics, and request IDs..
Alternatives
| Alternative | Use When | Don't Use When |
|---|---|---|
| Alternate framework in this cookbook | Team standard or existing monolith | Greenfield API with different constraints |
| Managed BaaS | CRUD-only MVP | Custom auth, workflows, or compliance needs |
| gRPC | Internal high-performance RPC | Public HTTP clients and browser access |
FAQs
When should I adopt Alembic migrations?
Use it when the patterns and trade-offs on this page match your API or data boundary.
What is the top production mistake with Alembic migrations?
Skipping validation, timeouts, or explicit error contracts at the HTTP edge.
How do I test Alembic migrations?
Use the framework test client, override dependencies, and assert status plus JSON shape.
Does Alembic migrations work with Python 3.14?
Yes - examples target Python 3.14 with pinned framework versions from the stack footer.
How does Alembic migrations relate to Pydantic 2?
Validate and serialize at boundaries; keep services working with typed domain objects.
Sync or async?
Prefer async routes when I/O dominates; keep CPU work small or offload to workers.
Where should business logic live?
Thin handlers; services own rules; repositories own queries.
How do I document APIs?
Publish OpenAPI or schema docs that match response models in code.
How do I handle versioning?
Explicit URL or header versioning with deprecation windows - avoid silent breaks.
What should I read next?
Follow the Related links for the next layer of depth in this section.
How do I stay secure?
Authenticate callers, authorize per resource, rate-limit, and never log secrets.
Performance first step?
Measure DB and upstream latency before swapping frameworks.
Related
- Databases Basics - Connections and transactions
- SQLAlchemy ORM 2.0 - ORM patterns
- Migrations with Alembic - Schema changes
- Async Database Access - asyncpg
Stack versions: This page was written for Python 3.14.0 (stable 3.14, maintenance 3.13), FastAPI 0.115+, Django 5.2, Flask 3.1, Pydantic 2, PyTorch 2.6+, pandas 2.2+, Polars 1.x, ruff 0.9+, and uv 0.6+.