ezmig 0.1.0b4

EZMig (Easy Migration) is a lightweight, SQL-first database migration tool designed for simplicity and CI/CD.

EZMig (Easy Migration)

EZMig is a lightweight, SQL-first database migration tool designed for simplicity and CI/CD.

It is:

• simple
• deterministic
• database-agnostic
• language-agnostic (just SQL)
• easy to integrate into pipelines

Documentation: https://klinvesta.github.io/ezmig/

---

✨ Philosophy

EZMig follows a simple rule:

SQL is the source of truth.

No ORM. No schema diffing. No magic.

Just:

• versioned SQL files
• applied in order
• tracked safely

---

🚀 Features

• ✅ Plain SQL migrations
• ✅ Linear, predictable execution
• ✅ Checksum validation (detect drift)
• ✅ Repeatable migrations (views, functions)
• ✅ Transaction support (when available)
• ✅ Rollback support via --ezmig:rollback
• ✅ Migration status tracking (pending, applied, changed…)
• ✅ Replay / force apply for dev iteration
• ✅ CLI-first design (CI/CD ready)
• ✅ Migration scaffolding via ezmig new

---

🗄️ Supported Databases

Database	Driver	Install	Example
SQLite	Built-in	uv tool install ezmig	examples/sqlite
PostgreSQL	psycopg2	uv tool install 'ezmig[postgres]'	examples/postgres
Oracle	oracledb	uv tool install 'ezmig[oracle]'	examples/oracle

Install just what you need:

• uv tool install ezmig — SQLite support only
• uv tool install 'ezmig[postgres]' — PostgreSQL support
• uv tool install 'ezmig[oracle]' — Oracle support
• uv tool install 'ezmig[all]' — All databases
• uvx ezmig --help — run without installing

---

📦 Installation

Linux one-line install

curl -fsSL https://raw.githubusercontent.com/klinvesta/ezmig/main/scripts/install.sh | BIN_DIR=$HOME/.local/bin sh

Standalone binary (no Python required)

Download a pre-built binary from the releases page.

uv tool install

uv tool install ezmig
uv tool install 'ezmig[postgres]'
uv tool install 'ezmig[oracle]'
uv tool install 'ezmig[all]'

uvx (run without installing)

uvx ezmig --help
uvx --from 'ezmig[postgres]' ezmig status --database dev
uvx --from 'ezmig[oracle]' ezmig plan --database dev

pipx

pipx install ezmig

pip

pip install ezmig                  # SQLite only
pip install 'ezmig[postgres]'      # + PostgreSQL
pip install 'ezmig[oracle]'        # + Oracle
pip install 'ezmig[all]'           # everything

See the Installation docs for full platform instructions and local binary builds.

---

🧰 Usage

ezmig new "create users"   # create versioned migration template
ezmig new "users_v" -r     # create repeatable migration template
ezmig status --database dev    # show current state for target
ezmig plan --database dev      # show pending migrations for target
ezmig apply --database dev     # apply migrations to target
ezmig apply --force --allow-replay --database dev   # force apply all (dev)
ezmig replay --migration "202601*.sql" --allow-replay --database dev
ezmig replay --repeatable "*.sql" --allow-replay --database dev
ezmig rollback --database dev  # rollback last migration on target

# inspect configured targets
ezmig config list
ezmig config show dev
ezmig config validate

---

📁 Project structure

migrations/
├── versioned/
│   ├── 20260101000000_init.sql
│   └── 20260315093000_add_users.sql
└── repeatable/
    ├── my_function.sql
    └── my_view.sql

---

📝 Migration format

Versioned migration

-- ezmig:apply
CREATE TABLE users (
  id SERIAL PRIMARY KEY,
  email TEXT NOT NULL UNIQUE
);

-- ezmig:rollback
DROP TABLE users;

• -- ezmig:apply → applied during ezmig apply
• -- ezmig:rollback → used during rollback (ezmig rollback)
• Required for rollback support

---

Repeatable migration

CREATE OR REPLACE VIEW active_users AS
SELECT * FROM users WHERE active = true;

---

🔢 Versioning

Versioned migrations

ezmig imposes no required filename format. Versioned migrations are sorted lexicographically by filename and executed in that order. You choose the naming convention.

Recommended convention: timestamps

For teams with more than one developer, the strongly recommended approach is a timestamp prefix:

20260323093000_add_payments_table.sql
20260323094512_add_audit_log.sql

Use YYYYMMDDHHmmss (14 digits). This works well because:

• Timestamps are globally unique per developer per second with no coordination needed.
• They sort correctly lexicographically, which is exactly how ezmig orders files.
• No two developers on separate branches will accidentally pick the same prefix.
• Merging branches never causes ordering conflicts — each migration keeps its natural position in time.

Why not sequential numbers? 001, 002, 003 only work on a single branch. As soon as two developers independently create 003_foo.sql and 003_bar.sql on different branches, you have a conflict with no safe resolution. Timestamps eliminate this class of problem entirely.

Execution order

Files are sorted lexicographically by filename and applied strictly in that order. With a timestamp prefix, lexicographic order equals chronological order, so the behaviour is intuitive.

⚠️ Never rename or edit a versioned migration file once it has been applied. ezmig stores the checksum of each file in ezmig_migrations. A rename looks like a missing file; a content change looks like drift. Both cause an error.

Immutability

Once a versioned migration has been applied, it is immutable:

• Its checksum is recorded in ezmig_migrations.
• If the file content changes, ezmig detects the mismatch during validation and refuses to proceed.
• To modify the schema after the fact, add a new migration with a later timestamp.

Repeatable migrations

Repeatable migrations have no ordering constraints. They are:

• Identified by their filename (used as the primary key in ezmig_migrations).
• Tracked by checksum — ezmig re-applies them whenever their content changes.
• Always applied after all versioned migrations in a given ezmig apply run.
• Sorted alphabetically among themselves.

Repeatable migrations are ideal for objects that can be safely recreated: views, functions, and stored procedures.

---

📊 Migration states

ezmig classifies migrations internally:

• PENDING → not yet applied
• APPLIED → already executed
• CHANGED → repeatable updated
• UNCHANGED → repeatable unchanged
• UNKNOWN → no adapter / offline mode

These states power both:

ezmig plan
ezmig status

---

🧾 Migration table

ezmig tracks applied migrations using:

CREATE TABLE ezmig_migrations (
  version      VARCHAR PRIMARY KEY,
  type         VARCHAR NOT NULL,
  checksum     VARCHAR NOT NULL,
  applied_at   TIMESTAMP NOT NULL,
  execution_ms INTEGER
);

---

🔄 Rollback

Rollback is explicit and safe:

ezmig rollback        # rollback last migration
ezmig rollback -n 3   # rollback last 3 migrations

• Executes -- ezmig:rollback blocks
• Runs in reverse order
• Wrapped in transactions (when supported)

---

🔁 Replay / force apply (dev)

For iterative local development, you can force re-apply migrations.

ezmig replay --migration "202601*.sql" --allow-replay --database dev
ezmig replay --repeatable "*.sql" --allow-replay --database dev
ezmig replay --all --allow-replay --database dev

# alias to replay everything in the target profile
ezmig apply --force --allow-replay --database dev

Safety gates:

• replay/force is disabled by default
• enable per-target with allow_replay = true in ezmig.toml, or pass --allow-replay

---

🔍 Validation

Before applying migrations, ezmig checks:

• checksum integrity
• missing migrations
• order consistency

Use it directly:

ezmig validate

ezmig validate exits with:

• 0 when validation passes
• 1 when validation fails

Validation fails include:

• applied versioned migration file missing on disk
• checksum drift on applied versioned migration
• out-of-order applied versioned migration history

Typical failure output:

Validation failed:
- Applied migration missing on disk: 20260101000000_init.sql
- Checksum drift detected for migration: 20260102000000_add_age.sql

Repeatable migration checksum changes are reported as warnings (they are expected to be re-applied on ezmig apply).

---

⚙️ Configuration

ezmig.toml:

[default]
database = "dev"

[migration.default]
versioned = "./migrations/versioned"
repeatable = [
  "./migrations/repeatable",
  "./migrations/views",
  "./migrations/packages",
  "./migrations/gtt",
]

[database.dev]
url = "postgresql://user:pass@localhost:5432/db"
migration = "default"
categories = { env = "dev", location = "local" }

Target selection options for DB commands:

• --database <name>
• --category <key=value> (repeatable)
• --group <name>

See docs/configuration.md for full schema and examples.

---

🔄 CI/CD Example

ezmig validate --database prod
ezmig plan --database prod
ezmig apply --database prod

Fail-fast CI pattern:

ezmig validate --database prod && ezmig apply --database prod

---

📚 Documentation

Published docs: https://klinvesta.github.io/ezmig/

Project documentation is organized under docs/ using MkDocs.

Build docs locally:

uv sync --group docs
uv run --group docs mkdocs serve

---

🧱 Design goals

• Keep the core small
• Avoid abstraction layers
• Favor explicit SQL over generated code
• Make failures obvious and safe

---

⚠️ Non-goals

ezmig intentionally does NOT:

• ❌ generate schema diffs
• ❌ manage ORM models
• ❌ support complex migration graphs

---

Why EZMig?

Because schema evolution should be:

• incremental
• continuous
• and quietly effective

---

📜 License

MIT