MongoMig

Alembic-style schema evolution and migrations for MongoDB.

Change your models, see exactly what changed, generate a migration, review its impact, and run it safely. Built for Python services (FastAPI, Flask, workers) on PyMongo or Beanie.

$ mongomig diff
USERS
  + status: string = 'active'                    REQUIRES_DATA_MIGRATION  backfill existing documents with 'active'
  + age: int | null = None                       SAFE  defaults to None: existing documents need no backfill
  + given_name: string                           MANUAL_REVIEW  required, with no default: choose a value for existing documents
  - first_name                                   WARNING  removed from the model; existing data is kept (not deleted)
  + index users_email_unique (email ↑, unique)   WARNING  fails if existing documents contain duplicates

Possible rename: users.first_name → given_name (85% similar). If so, pass --rename users.first_name:given_name

$ mongomig revision --autogenerate -m "evolve users" --rename users.first_name:given_name
Generated b91baf4fd592 → migrations/versions/20260925_1115_b91baf4fd592_evolve_users.py

$ mongomig plan
b91baf4fd592  evolve users   Risk: HIGH
  users  rename_field  $rename first_name           ~4,218,901 docs · collection scan
  users  backfill      $set status                  ~4,218,901 docs · collection scan
  users  create_index  users_email_unique (unique)  4,218,901 docs
         ⚠ will fail: duplicate values exist, e.g. {'email': 'sam@example.com'}

$ mongomig upgrade

Why MongoMig

• Autogenerate from your models: plain Pydantic (@collection) or Beanie documents, no extra declarations. Backfills with your defaults, renames, indexes, $jsonSchema validators.
• Every change is classified: SAFE, WARNING, REQUIRES_DATA_MIGRATION or MANUAL_REVIEW. Risky changes become commented TODO(review) blocks, and data is never deleted automatically.
• Deterministic diffs: models are compared with a committed schema snapshot rather than a sampled database, so results are the same on every machine and in CI.
• Knows how you store data: model_dump() vs jsonable_encoder stores dates as different BSON types, and MongoMig warns about types PyMongo can't store.
• Production-grade execution: distributed lock, batched operations with progress and ETA, retries, checkpoints so a failed migration resumes where it stopped (mongomig resume), transactions, checksums, confirmation before destructive migrations, and restorable backups.
• See the impact first: plan / --dry-run estimate documents touched and collection scans, and predict unique-index failures and validator rejections against real data.
• Catch drift: mongomig drift compares your models with the data actually stored. It reports missing fields, unexpected types (age: string in 5.7% of documents), unknown fields, and missing indexes or validators, with thresholds for CI or scheduled checks.
• Keeps history manageable: mongomig squash replaces old revisions with one. Existing databases adopt it automatically, and new ones replay a single revision.
• Built for CI: mongomig validate, --json output everywhere, documented exit codes.

Install

pip install mongomig            # add [beanie] for Beanie support

Python 3.11+, MongoDB 6.0+ (tested on 6.0, 7.0, 8.0).

Quick start

mongomig init                                   # mongomig.yaml + migrations/
export MONGODB_URI="mongodb://localhost:27017"
# register your models in migrations/env.py (see below), then:
mongomig revision --autogenerate -m "initial"   # new project
mongomig baseline                               # ...or an existing database
mongomig upgrade

# app/models.py
from pydantic import BaseModel
from mongomig import collection, Index

@collection("users", indexes=[Index("email", unique=True)])
class User(BaseModel):
    name: str
    email: str
    status: str = "active"

# migrations/env.py
from mongomig import MongoMetadata

import app.models  # noqa: F401

target_metadata = MongoMetadata.default(storage="python")   # or "json"; Beanie: register_beanie(...)

Everyday loop: edit models → mongomig diff → mongomig revision --autogenerate -m "..." → review and commit → CI mongomig validate → deploy mongomig plan + mongomig upgrade. On a schedule: mongomig drift --check.

$ mongomig drift
USERS  10,000 sampled of ~4,982,133 documents
  ✗ (index)           index users_email_unique (email ↑, unique) is not in the database
  ✗ age               expected int | null · observed string 5.70%  > 0.5%
  ✗ profile.verified  required by the model · missing in 10.1% of documents  > 0.5%
  ! nickname          not in the model · present in 0.36% of documents  ≤ 1%

Documentation

Full docs: https://anon-000.github.io/mongomig/

• Getting started
• Recipes: step-by-step guides for renames, type changes, unique indexes, team merges, deploys, FastAPI, testing, rollbacks and more
• Concepts: snapshots, storage profiles, classifications
• Writing migrations: ctx and ctx.ops reference
• Autogenerate: what gets generated and how to review it
• Production guide: permissions, plan, locking, backups, recovery
• Coming from Alembic or hand-written scripts?
• CLI reference · Python API · Troubleshooting

Examples: FastAPI + Pydantic · FastAPI + Beanie · FastAPI with migration tests

Status

The MVP is complete (config, revisions, upgrade/downgrade, schema inspection, diff, autogenerate, plan/dry-run, locking, batching, backups, CI checks), plus drift detection, resumable migrations, transactions and squashing. Next: observability (OpenTelemetry / Prometheus metrics). Ideas and feedback welcome.

Development

make install        # uv venv + editable install with dev extras
make mongo-up       # MongoDB 7 single-node replica set in Docker
make check          # ruff + mypy --strict + pytest

Integration tests skip when MongoDB isn't running. To test another MongoDB version, run MONGO_VERSION=8.0 make mongo-up.

Releasing

1. Bump src/mongomig/_version.py and update CHANGELOG.md, then merge to main.
2. Create a GitHub Release with tag v<version> and publish it. The release workflow checks the version, builds and smoke-tests the wheel, and publishes to PyPI via Trusted Publishing.

Contributing

Bug reports, ideas and pull requests are welcome. See CONTRIBUTING.md.

License

MIT