Shared from "Study" on Inkdown
Alembic is Prisma Migrate for Python/SQLAlchemy. You change your models → Alembic generates the SQL → you apply it to the DB.
The core problem it solves: your Python model changed, but the real database has no idea. Alembic bridges that gap with versioned, reproducible migration files.
| Concept | Prisma | Drizzle | Alembic |
|---|---|---|---|
| Define schema |
schema.prisma |
schema.ts |
SQLAlchemy models (models/) |
| Generate migration | prisma migrate dev | drizzle-kit generate | alembic revision --autogenerate |
| Apply migration | prisma migrate deploy | drizzle-kit push | alembic upgrade head |
| Migration files | migrations/ | migrations/ | alembic/versions/ |
| Roll back | ❌ not built-in | ❌ not built-in | ✅ alembic downgrade -1 |
Alembic's rollback is a genuine advantage over Prisma/Drizzle. Every migration has an
upgrade()anddowngrade()— you can go backwards cleanly.
In Prisma, the .prisma file is the single source of truth — it generates both the client and migrations.
In SQLAlchemy + Alembic, they are two separate tools:
You have to manually wire them together in alembic/env.py.
⚠️ #1 Gotcha: If you add a new model file and forget to import it here, Alembic will not detect it — and won't generate a migration for it.
Alembic compares your current models vs what the DB looks like right now and generates a migration file automatically:
create_all | Alembic | |
|---|---|---|
| Use case | Local dev only | Staging + Production |
| Tracks changes | ❌ No | ✅ Yes, versioned |
| Safe on existing data | ❌ Won't alter existing tables | ✅ Generates precise ALTER statements |
| Rollback | ❌ No | ✅ Yes, via downgrade() |
Rule of thumb: use
create_allto spin up a fresh local DB fast. Use Alembic for anything that has real data.
Always write a proper
downgrade(). It's what makes rollbacks possible.
| Gotcha | Fix |
|---|---|
| New model not detected | Import it in alembic/env.py |
--autogenerate misses some changes | Always review generated files before applying |
| Rolling back drops columns | Data in those columns is gone — back up first |
Running create_all alongside Alembic | Don't mix them — pick one approach per environment |