Using modl? Check out modl-migrate.
- Usable as a CLI tool or as a library
- Supports SQLite, PostgreSQL, MySQL, MSSQL and Oracle databases (through gorp)
- Can embed migrations into your application
- Migrations are defined with SQL for full flexibility
- Atomic migrations
- Up/down migrations to allow rollback
- Supports multiple database types in one project
- Works great with other libraries such as sqlx
To install the library and command line program, use the following:
go get -v github.com/rubenv/sql-migrate/...
$ sql-migrate --help
usage: sql-migrate [--version] [--help] <command> [<args>]
Available commands are:
down Undo a database migration
new Create a new migration
redo Reapply the last migration
status Show migration status
up Migrates the database to the most recent version available
Each command requires a configuration file (which defaults to dbconfig.yml
, but can be specified with the -config
flag). This config file should specify one or more environments:
development:
dialect: sqlite3
datasource: test.db
dir: migrations/sqlite3
production:
dialect: postgres
datasource: dbname=myapp sslmode=disable
dir: migrations/postgres
table: migrations
(See more examples for different set ups here)
Also one can obtain env variables in datasource field via os.ExpandEnv
embedded call for the field.
This may be useful if one doesn't want to store credentials in file:
production:
dialect: postgres
datasource: host=prodhost dbname=proddb user=${DB_USER} password=${DB_PASSWORD} sslmode=required
dir: migrations
table: migrations
The table
setting is optional and will default to gorp_migrations
.
The environment that will be used can be specified with the -env
flag (defaults to development
).
Use the --help
flag in combination with any of the commands to get an overview of its usage:
$ sql-migrate up --help
Usage: sql-migrate up [options] ...
Migrates the database to the most recent version available.
Options:
-config=dbconfig.yml Configuration file to use.
-env="development" Environment.
-limit=0 Limit the number of migrations (0 = unlimited).
-dryrun Don't apply migrations, just print them.
The new
command creates a new empty migration template using the following pattern <current time>-<name>.sql
.
The up
command applies all available migrations. By contrast, down
will only apply one migration by default. This behavior can be changed for both by using the -limit
parameter.
The redo
command will unapply the last migration and reapply it. This is useful during development, when you're writing migrations.
Use the status
command to see the state of the applied migrations:
$ sql-migrate status
+---------------+-----------------------------------------+
| MIGRATION | APPLIED |
+---------------+-----------------------------------------+
| 1_initial.sql | 2014-09-13 08:19:06.788354925 +0000 UTC |
| 2_record.sql | no |
+---------------+-----------------------------------------+
You can see how to run setups for different setups by executing the .sh
files in test-integration
# Run mysql-env.sh example (you need to be in the project root directory)
./test-integration/mysql-env.sh
If you are using MySQL, you must append ?parseTime=true
to the datasource
configuration. For example:
production:
dialect: mysql
datasource: root@/dbname?parseTime=true
dir: migrations/mysql
table: migrations
See here for more information.
Import sql-migrate into your application:
import "github.com/rubenv/sql-migrate"
Set up a source of migrations, this can be from memory, from a set of files or from bindata (more on that later):
// Hardcoded strings in memory:
migrations := &migrate.MemoryMigrationSource{
Migrations: []*migrate.Migration{
&migrate.Migration{
Id: "123",
Up: []string{"CREATE TABLE people (id int)"},
Down: []string{"DROP TABLE people"},
},
},
}
// OR: Read migrations from a folder:
migrations := &migrate.FileMigrationSource{
Dir: "db/migrations",
}
// OR: Use migrations from a packr box
migrations := &migrate.PackrMigrationSource{
Box: packr.NewBox("./migrations"),
}
// OR: Use migrations from bindata:
migrations := &migrate.AssetMigrationSource{
Asset: Asset,
AssetDir: AssetDir,
Dir: "migrations",
}
Then use the Exec
function to upgrade your database:
db, err := sql.Open("sqlite3", filename)
if err != nil {
// Handle errors!
}
n, err := migrate.Exec(db, "sqlite3", migrations, migrate.Up)
if err != nil {
// Handle errors!
}
fmt.Printf("Applied %d migrations!\n", n)
Note that n
can be greater than 0
even if there is an error: any migration that succeeded will remain applied even if a later one fails.
Check the GoDoc reference for the full documentation.
Migrations are defined in SQL files, which contain a set of SQL statements. Special comments are used to distinguish up and down migrations.
-- +migrate Up
-- SQL in section 'Up' is executed when this migration is applied
CREATE TABLE people (id int);
-- +migrate Down
-- SQL section 'Down' is executed when this migration is rolled back
DROP TABLE people;
You can put multiple statements in each block, as long as you end them with a semicolon (;
).
You can alternatively set up a separator string that matches an entire line by setting sqlparse.LineSeparator
. This
can be used to imitate, for example, MS SQL Query Analyzer functionality where commands can be separated by a line with
contents of GO
. If sqlparse.LineSeparator
is matched, it will not be included in the resulting migration scripts.
If you have complex statements which contain semicolons, use StatementBegin
and StatementEnd
to indicate boundaries:
-- +migrate Up
CREATE TABLE people (id int);
-- +migrate StatementBegin
CREATE OR REPLACE FUNCTION do_something()
returns void AS $$
DECLARE
create_query text;
BEGIN
-- Do something here
END;
$$
language plpgsql;
-- +migrate StatementEnd
-- +migrate Down
DROP FUNCTION do_something();
DROP TABLE people;
The order in which migrations are applied is defined through the filename: sql-migrate will sort migrations based on their name. It's recommended to use an increasing version number or a timestamp as the first part of the filename.
Normally each migration is run within a transaction in order to guarantee that it is fully atomic. However some SQL commands (for example creating an index concurrently in PostgreSQL) cannot be executed inside a transaction. In order to execute such a command in a migration, the migration can be run using the notransaction
option:
-- +migrate Up notransaction
CREATE UNIQUE INDEX people_unique_id_idx CONCURRENTLY ON people (id);
-- +migrate Down
DROP INDEX people_unique_id_idx;
Embedding migrations with packr
If you like your Go applications self-contained (that is: a single binary): use packr to embed the migration files.
Just write your migration files as usual, as a set of SQL files in a folder.
Use the PackrMigrationSource
in your application to find the migrations:
migrations := &migrate.PackrMigrationSource{
Box: packr.NewBox("./migrations"),
}
If you already have a box and would like to use a subdirectory:
migrations := &migrate.PackrMigrationSource{
Box: myBox,
Dir: "./migrations",
}
Embedding migrations with bindata
As an alternative, but slightly less maintained, you can use bindata to embed the migration files.
Just write your migration files as usual, as a set of SQL files in a folder.
Then use bindata to generate a .go
file with the migrations embedded:
go-bindata -pkg myapp -o bindata.go db/migrations/
The resulting bindata.go
file will contain your migrations. Remember to regenerate your bindata.go
file whenever you add/modify a migration (go generate
will help here, once it arrives).
Use the AssetMigrationSource
in your application to find the migrations:
migrations := &migrate.AssetMigrationSource{
Asset: Asset,
AssetDir: AssetDir,
Dir: "db/migrations",
}
Both Asset
and AssetDir
are functions provided by bindata.
Then proceed as usual.
Adding a new migration source means implementing MigrationSource
.
type MigrationSource interface {
FindMigrations() ([]*Migration, error)
}
The resulting slice of migrations will be executed in the given order, so it should usually be sorted by the Id
field.
Usage with sqlx
This library is compatible with sqlx. When calling migrate just dereference the DB from your *sqlx.DB
:
n, err := migrate.Exec(db.DB, "sqlite3", migrations, migrate.Up)
// ^^^ <-- Here db is a *sqlx.DB, the db.DB field is the plain sql.DB
if err != nil {
// Handle errors!
}
This library is distributed under the MIT license.