Automigrate

Auto-generated database schema migrations for Clojure. Define models as plain EDN data and create database schema migrations automatically based on changes to the models.

Features

• declaratively define db schema as models in EDN;
• create migrations automatically based on model changes;
• migrate db schema in forward and backward directions;
• manage migrations for: tables, indexes, constraints, enum types;
• view actual SQL or human-readable description for a migration;
• optionally add a custom SQL migration for specific cases;
• use with PostgreSQL :information_source: [other databases are planned] .

Quick overview

https://github.com/abogoyavlensky/automigrate/assets/1375411/880db134-f2ed-46b4-9e77-72e326b6bf56

State

Project is in alpha state till the 1.0.0 version and is not yet ready for production use. Breaking changes are possible.

Usage

Installation

tools.deps -X option

A config for development environment could look like the following:

deps.edn

{...
 :aliases {...
           :migrations {:extra-deps {net.clojars.abogoyavlensky/automigrate {:mvn/version "<LATEST VERSION>"}
                                     org.postgresql/postgresql {:mvn/version "42.3.1"}}
                        :ns-default automigrate.core
                        :exec-args {:models-file "resources/db/models.edn"
                                    :migrations-dir "resources/db/migrations"
                                    :jdbc-url "jdbc:postgresql://localhost:5432/mydb?user=myuser&password=secret"}}}}

:information_source: You can move postgres driver to project's :deps section. You can choose any paths you want for :models-file and :migrations-dir.

Then you could use it as:

clojure -X:migrations make

Leiningen

:construction: Leiningen support is planned.

Getting started

After configuration, you are able to create models.edn file with first model, make migration for it and migrate db schema. Paths for migrations and models can be chosen as you want. A model is the representation of a database table which is described in EDN structure. Let's do it step by step.

Add model

resources/db/models.edn

{:book [[:id :serial {:primary-key true}]
        [:name [:varchar 255] {:null false}]
        [:description :text]]}

Make migration

$ clojure -X:migrations make
Created migration: resources/db/migrations/0001_auto_create_table_book.edn
Actions:
  - create table book

Migration can contain multiple actions. Every action is converted to a SQL query. The migration at resources/db/migrations/0001_auto_create_table_book.edn will look like:

({:action :create-table,
  :model-name :book,
  :fields
  {:id {:primary-key true, :type :serial},
   :name {:null false, :type [:varchar 255]},
   :description {:type :text}}})

Migrate

Existing migrations will be applied one by one in order of migration number:

$ clojure -X:migrations migrate
Applying 0001_auto_create_table_book...
0001_auto_create_table_book successfully applied.

That's it. In the database you can see a newly created table called book with defined columns and one entry in table automigrate_migrations with new migration 0001_auto_create_table_book.

List and explain migrations

To view status of existing migrations you can run:

$ clojure -X:migrations list
Existing migrations:
[x] 0001_auto_create_table_book.edn

To view raw SQL for existing migration you can run command explain with appropriate number:

$ clojure -X:migrations explain :number 1
SQL for forward migration 0001_auto_create_table_book.edn:

BEGIN;

CREATE TABLE book (
  id SERIAL PRIMARY KEY,
  name VARCHAR(255) NOT NULL,
  description TEXT
);

COMMIT;

All SQL queries of the migration are wrapped by a transaction.

:information_source: For a slightly more complex example please check models.edn and README.md from the examples dir of this repo.

Documentation

Model definition

Models are represented as a map with the model name as a keyword key and the value describing the model itself. A model's definition could be a vector of vectors in the simple case of just defining fields. As we saw in the previous example:

{:book [[:id :serial {:primary-key true}]
        [:name [:varchar 255] {:null false}]
        [:description :text]]}

Or it could be a map with keys :fields, :indexes (optional) and :types (optional). Each of these is also a vector of vectors. The same model from above could be described as a map:

{:book {:fields [[:id :serial {:primary-key true}]
                [:name [:varchar 255] {:null false}]
                [:description :text]]}}

Fields

Each field is a vector of three elements: [:field-name :field-type {:some-option :option-value}]. The third element is optional, but name and type are required.

The first element is the name of a field and must be a keyword.

Field types

The second element could be a keyword or a vector of keyword and params. Available field types are matched with PostgreSQL built-in data types and presented in the following table:

Field type	Description
:integer
:smallint
:bigint
:float
:real
:serial	Auto-incremented pg integer field.
:bigserial	Auto-incremented pg bigint field.
:smallserial	Auto-incremented pg serial2 field.
:numeric or [:numeric <pos-int>? <int>]	Numeric type with optional precision and scale params. Default value could be set as numeric string, bigdec, float, int and nil: "10.22", 10.22M, 10, 10.22, nil.
:decimal or [:decimal <pos-int>? <int>]	Numeric type with optional precision and scale params. Same as :numeric.
:uuid
:boolean
:text
:time or [:time <int>]
:timetz or [:timetz <int>]
:timestamp or [:timestamp <int>]
:timestamptz or [:timestamptz <int>]
:interval or [:interval <int>]
:date
:point
:json
:jsonb
:varchar or [:varchar <pos-int>]	Second element is the length of value.
:char or [:char <pos-int>]	Second element is the length of value.
:float or [:float <pos-int>]	Second element is the minimum acceptable precision in binary digits.
[:enum <enum-type-name>]	To use enum type you should define it in :types section in model.
:box
:bytea
:cidr
:circle
:double-precision
:inet
:line
:lseg
:macaddr
:macaddr8
:money
:path
:pg_lsn
:pg_snapshot
:polygon
:tsquery
:tsvector
:txid_snapshot
:xml
:bit or [:bit <pos-int>]
:varbit or [:varbit <pos-int>]

Doc reference to the PostgreSQL built-in general-purpose data types: https://www.postgresql.org/docs/current/datatype.html#DATATYPE-TABLE

Notes

• <...>? - param is optional.
• or - an alternative definition of type.

:information_source: There are fixed field types because automigrate validates type of field and default value to have errors as early as possible before running migration against database.

Field options

Options value is a map where key is the name of the option and value is the available option value. Available options are presented in the table below:

Field option	Description	Required?	Value
:null	Set to false for non-nullable field. Field is nullable by default if the option is not set.	false	boolean?
:primary-key	Set to true for making primary key field.	false	true?
:unique	Set to true to add unique constraint for a field.	false	true?
:default	Default value for a field.	false	boolean?, integer?, float?, decimal?, string?, nil?, or fn defined as [:keyword <integer? or float? or string?>]
:foreign-key	Set to namespaced keyword to point to a primary key field from another model.	false	:another-model/field-name
:on-delete	Specify delete action for :foreign-key.	false	:cascade, :set-null, :set-default, :restrict, :no-action
:on-update	Specify update action for :foreign-key.	false	:cascade, :set-null, :set-default, :restrict, :no-action
:check	Set condition in Honeysql format to create custom CHECK for a column.	false	Example: [:and [:> :month 0] [:<= :month 12]]
:array	Can be added to any field type to make it array.	false	string?, examples: "[]", "[][]", [][10][3]
:comment	Add a comment on the field.	false	string?

Indexes

Each index is a vector of three elements: [:name-of-index :type-of-index {:fields [:field-from-model-to-index] :unique boolean? :where [...]}] Name, type and :fields in options are required.

The first element is the name of an index and must be a keyword.

Index types

The second element is an index type and must be a keyword of available index types:

Field type
:btree
:gin
:gist
:spgist
:brin
:hash

Index options

The options value is a map where key is the name of the option and value is the available option value. The option :fields is required, others are optional.
Available options are presented in the table below:

Field option	Description	Required?	Value
:fields	Vector of fields as keywords. Index will be created for those fields.	true	[:field-name ...]
:unique	Set to true if index should be unique.	false	true?
:where	Set condition in Honeysql format to create partial index.	false	Example: [:> amount 10]

Types

:information_source: At the moment only Enum type is supported.

Each type is a vector of three elements: [:name-of-type :type-of-type {...}] Name, type-of-type and options are required.

The first element is the name of a type and must be a keyword.

Type of type

The second element is a type of type and must be a keyword of available types:

Field type
:enum

Enum type

Each enum type is a vector of three elements: [:name-of-type :enum {:choices [<str>]}]

Options for enum type must contain the :choices value with vector of strings. :choices represent enum values for the type.

An example of model definition with enum type:

{:account {:fields [[:id :serial]
                    [:role [:enum :account-role]]]
           :types [[:account-role :enum {:choices ["admin" "customer"]}]]}}

Limitations:

• :choices can't be empty;
• values in :choices must be unique for the particular type;
• removing a value from :choices of existing type is not supported;
• re-ordering values in :choices of existing type is not supported;

CLI interface

Available commands are: make, migrate, list, explain, help. Let's see them in detail by section.

:information_source: Assume that args :models-file, :migrations-dir and :jdbc-url are supposed to be set in deps.edn alias.

Common args for all commands:

Argument	Description	Required?	Possible values	Default value
:models-file	Path to models file.	true (only for make)	string path (example: "path/to/models.edn")	not provided
:migrations-dir	Path to store migrations dir.	true	string path (example: "path/to/migrations")	not provided
:jdbc-url	Database connection defined as JDBC-url.	true (only for migrate and list)	string jdbc url (example: "jdbc:postgresql://localhost:5432/mydb?user=myuser&password=secret")	not provided
:migrations-table	Model name for storing applied migrations.	false	string (example: "migrations")	"automigrate_migrations"

make

Create migration for new changes in models file. It detects the creating, updating and deleting of tables, columns and indexes. Each migration is wrapped by transaction by default.

Specific args:

Argument	Description	Required?	Possible values	Default value
:type	Type of migration file.	false	:empty-sql	not provided, migration will be created automatically
:name	Custom name for migration file separated by underscores.	false (:warning: required for :empty-sql type)	string (example: "add_custom_trigger")	generated automatically by first migration action

Examples

Create migration automatically with auto-generated name:

$ clojure -X:migrations :make
Created migration: resources/db/migrations/0001_auto_create_table_book.edn
Actions:
  ...

Create migration automatically with custom name:

$ clojure -X:migrations make :name create_table_author
Created migration: resources/db/migrations/0002_create_table_author.edn
Actions:
  ...

Create empty SQL migration with custom name:

$ clojure -X:migrations make :type :empty-sql :name add_custom_trigger
Created migration: resources/db/migrations/0003_add_custom_trigger.sql

Try to create migration without new changes in models:

$ clojure -X:migrations make
There are no changes in models.

migrate

Applies change described in migration to database. Applies all unapplied migrations by number order if arg :number is not presented in command. Throws error for same migration number.

Backward migration is fully implemented. For auto-generated and SQL migrations, it is possible to revert migration and to delete appropriate entry from migrations table. Database changes will be reverted.

In forward direction if specified migration :number is included, meaning if, for example, :number 3 the migration with number 3 will be applied. In backward migration the :number is excluded, so all migrations until the specified number will be reverted but not the target one. For instance if we have 3 migrations as applied, and want to revert just the 3d and 2d ones, we can run migrate command with :number 1. 3d and 3d migrations will be reverted, but the first one will stay applied.

Specific args:

Argument	Description	Required?	Possible values	Default value
:number	Number of migration which should be a target point. In forward direction, migration by number will by applied. In backward direction, migration by number will be reverted.	false	integer (example: 1 for migration 0001_...)	not provided, last migration number by default

Examples

Migrate forward all unapplied migrations:

$ clojure -X:migrations migrate
Appyling 0001_auto_create_table_book...
0001_auto_create_table_book successfully applied.
Appyling 0002_create_table_author...
0002_create_table_author successfully applied.
Appyling 0003_add_custom_trigger...
0003_add_custom_trigger successfully applied.

Migrate forward up to particular migration number (included):

$ clojure -X:migrations migrate :number 2
Appyling 0001_auto_create_table_book...
0001_auto_create_table_book successfully applied.
Appyling 0002_create_table_author...
0002_create_table_author successfully applied.

Migrate backward down to particular migration number (excluded):

$ clojure -X:migrations migrate :number 1
Reverting 0002_create_table_author...
0002_create_table_author successfully reverted.

Migrate backward to initial state of database:

$ clojure -X:migrations migrate :number 0
Reverting 0003_add_custom_trigger...
0003_add_custom_trigger successfully reverted.
Reverting 0002_create_table_author...
0002_create_table_author successfully reverted.
Reverting 0001_auto_create_table_book...
0001_auto_create_table_book successfully reverted.

Try to migrate already migrated migrations:

$ clojure -X:migrations migrate
Nothing to migrate.

Try to migrate up to not existing migration:

$ clojure -X:migrations migrate :number 10
-- ERROR -------------------------------------

Invalid target migration number.

list

Print out list of existing migrations with statuses displayed as boxes before migration name:

• [x] - applied;
• [ ] - not applied.

No specific args.

Examples:

View list of partially applied migrations:

$ clojure -X:migrations list
Existing migrations:
[x] 0001_auto_create_table_book.edn
[ ] 0002_create_table_author.edn
[ ] 0003_add_custom_trigger.sql

explain

Print out actual raw SQL for particular migration by number.

Specific args:

Argument	Description	Required?	Possible values	Default value
:number	Number of migration which should be explained.	true	integer (example: 1 for migration 0001_...)	not provided
:direction	Direction in which migration should be explained.	false	:forward, :backward	:forward
:format	Format of explanation.	false	:sql, :human	:sql

Examples:

View raw SQL for migration in forward direction:

$ clojure -X:migrations explain :number 1
SQL for forward migration 0001_auto_create_table_book.edn:

BEGIN;

CREATE TABLE book (
  id SERIAL PRIMARY KEY,
  name VARCHAR(255) NOT NULL,
  description TEXT
);

COMMIT;

View raw SQL for migration in backward direction:

$ clojure -X:migrations explain :number 1 :direction backward
SQL for backward migration 0001_auto_create_table_book.edn:

BEGIN;

DROP TABLE IF EXISTS book;

COMMIT;

help

You can print short doc info for a particular command or the tool itself by running help command.

Args:

Argument	Description	Required?	Possible values	Default value
:cmd	Command name.	false	make, migrate, list, explain, help	not provided, by default prints doc for all commands

Examples

Print doc for all available commands:

$ clojure -X:migrations help
Auto-generated database migrations for Clojure.

Available commands:
...

Print doc for a particular command:

$ clojure -X:migrations help :cmd make
Create a new migration based on changes to the models.

Available options:
...

Custom SQL migration

There are some specific cases which are not yet supported by auto-migrations. There are cases when you need to add simple data migration. You can add a custom SQL migration which contains raw SQL for forward and backward directions separately in single SQL-file. For that you can run the following command for making empty SQL migration with custom name:

$ clojure -X:migrations make :type :empty-sql :name make_all_accounts_active
Created migration: resources/db/migrations/0003_make_all_accounts_active.sql

The newly created file will look like:

-- FORWARD


-- BACKWARD

You can fill it with two block of queries for forward and backward migration. Backward migration block is not mandatory and can be empty. For example:

-- FORWARD

UPDATE account
SET is_active = true;

-- BACKWARD

UPDATE account
SET is_active = false;

Then migrate it as usual:

$ clojure -X:migrations migrate
Appyling 0003_make_all_accounts_active...
0003_make_all_accounts_active successfully applied.

Use in production

:warning: The library is not yet ready for production use. But it is really appreciated if you try it out! :wink:

For now, there is just a single way to configure db connection url as a :jdbc-url arg for a command. The idea here is that you could override default dev :jdbc-url value from deps.edn by running migrate command with env var using bash-script, makefile or whatever you want:

$ clojure -X:migrations migrate :jdbc-url $DATABASE_URL 
Appyling ...

The downside of that approach could be a lack of ability to use a common config for a project. In the future there could be more convenient options for configuration.

Roadmap

• [x] Enum type of fields.
• [x] All built-in data types.
• [x] Array data types.
• [x] Comment on field.
• [x] Partial indexes.
• [x] Auto-generated backward migration.
• [x] Field level CHECK constraints.
• [ ] Data-migration using Clojure.
• [ ] Support for SQLite.
• [ ] Model level constraints.
• [ ] Optimized auto-generated SQL queries.
• [ ] Visual representation of DB schema.
• [ ] Support for MySQL.

Things still in design

• How to handle common configuration conveniently?
• Should args :models-file and :migrations-dir are set by default?
• Should it be possible to set arg :jdbc-url as an env var?
• More consistent and helpful messages for users, maybe using fipp library.
• Ability to separate models by multiple files.
• Move transformations out of clojure spec conformers.
• Try to replace spec with malli.
• Simplify model definition just as map with key :type instead of vector of 3 items.
• Disable field types validation at all, or add ability to set arbitrary custom type.
• Handle of model/field renaming.

Inspired by

• Django Migrations
• Prisma Migrate

Thanks to projects

• Honey SQL
• Dependency
• Differ

Resources

• Announcing automigrate (blog post)

Development

Run locally

make up  # run docker-compose with databases for development
make repl  # run builtin repl with dev aliases; also you could use any repl you want
make test  # run whole tests locally against testing database started by docker-compose
make fmt  # run formatting in action mode
make lint  # run linting
make check-deps  # run checking new versions of deps in force mode

Release new version

make install-snapshot :patch  # build and install locally a new version of lib based on latest git tag and using semver
make deploy-snapshot :patch  # build and deploy to Clojars next snapshot version from local machine
make release :patch  # bump git tag version by semver rules and push to remote repo

License

Copyright © 2021 Andrey Bogoyavlenskiy

Distributed under the MIT License.