Skip to main content

Applying Migrations

With the atlas migrate apply command, users can apply pending migrations to database(s). The typical flow for introducing schema changes to databases is as follows: DevelopCheck (CI)Push (CD)Deploy.

  1. Develop - Generate a migration file with the desired database changes using the atlas migrate diff command.
  2. Check (CI) - Use atlas migrate lint to validate migrations, ensuring they don't conflict with other team members' changes and align with best practices. Add Atlas to your CI pipeline in GitHub Actions or GitLab to review migrations files before they get merged into the main branch.
  3. Push (Delivery) - Use atlas migrate push, or set up a CI pipeline to push the latest migrations state to Atlas Cloud. Alternatively, you can package the migrations directory into a custom Docker image and push it to an artifactory.
  4. Deploy - Use atlas migrate apply to apply the pending migrations to your database(s).

Flags and Arguments

By default, atlas migrate apply executes all pending migrations. However, you can pass an optional argument to limit the number of migrations applied. For instance, atlas migrate apply 2 will apply up to 2 pending migrations.

The following flags are required:

  • --url the URL to the database to apply migrations on.
  • --dir the URL to the migration directory. It defaults to file://migrations.
Reading remote directories

Users who have connected or pushed their migration directories to Atlas Cloud can read the migrations state directly from there without needing to have them locally. For example, atlas migrate apply --dir "atlas://app" will apply the pending migrations of the app project based on the most recent pushed state. To see it in action, run the following:

Login or signup:

atlas login

Push a local migration directory and name it app:

atlas migrate push app \
--dev-url "docker://postgres/15/dev?search_path=public"

Deploy to a local database the remote migration directory named app:

atlas migrate apply \
--dir "atlas://app" \
--url "postgres://postgres:pass@:5432/example?search_path=public&sslmode=disable"

Deploy a specific tag to a local database the remote migration directory named app:

atlas migrate apply \
--dir "atlas://app?tag=39e7e4e35fce7409bd26d25d8140061695d4ffd5" \
--url "postgres://postgres:pass@:5432/example?search_path=public&sslmode=disable"

Schema Revision Information

Atlas saves information about the applied migrations on a table called atlas_schema_revisions in the connected database schema (e.g. mysql://user@host/my_schema or postgres://user@host/db?search_path=my_schema). If the database connection is not bound to a specific schema (e.g. mysql://user@host/ or postgres://user@host/db), the table is stored in its own schema called atlas_schema_revisions. This behavior can be changed by setting the schema manually:

  • --revisions-schema my_schema to store the data in my_schema.atlas_schema_revisions.

Transaction Configuration

By default, Atlas creates one transaction per migration file and will roll back that transaction if a statement in the wrapped migration fails to execute. Atlas supports three different transaction modes:

  • --tx-mode file (default) will wrap each pending migration into its own transaction.
  • --tx-mode all will wrap all pending migration files into one transaction.
  • --tx-mode none will not create any transaction. If a statement fails, the execution will stop. However, Atlas is smart enough to detect which statement fails and on another migration attempt will continue with the failed statement. This means altering the migration file from the failed statements onwards is safe and recommended.

Please be aware, that non DDL transactional databases like MySQL (due to implicit commits) can not be safely rolled back completely, and you might end up with a mismatched schema and revision table state. Atlas will handle those cases in future releases. A good source of information can be found in the PostgreSQL wiki.

File level transaction mode

The --atlas:txmode directive can be used to override the transaction mode for a specific migration file:

-- atlas:txmode none

CREATE INDEX CONCURRENTLY name_idx ON users (name);

Execution Order

The --exec-order flag controls how Atlas computes and executes pending migration files to the database. Atlas supports three different order execution modes:

  • linear (default) - Atlas expects a linear history and fails if it encounters files that were added out of order. This option ensures files are executed in a consistent order, guaranteeing deterministic behavior. It can be enforced in CI using the atlas migrate lint command. Learn more about non-linear changes in the documentation.
  • linear-skip - This option is a softer version of linear, meaning that if Atlas encounters a new file that was not added in sequential order (its version is lower than the database version), it will be skipped.
  • non-linear - This option directs Atlas to execute migration files that were added out of order. Note, although this option can be useful in local development, it is strongly discouraged in real production environments. Executing files out of order cannot guarantee deterministic behavior and may lead to failures (e.g., conflicted migrations). Rolling back to a specific version of the migration directory might be challenging, as the state of the database could differ from the state of the directory.

Handling Out-of-Order Errors

Files were added out of order

You've encountered this issue because your database is at version Z, but there is a file(s) in your migration directory pending to be applied with version Y, where Y < Z. This indicates it was added out of order, as its version is lower than the current database version. Below are multiple options to resolve it depending on your environment:

  • Local environment (development): Developers might encounter this issue if they have a migration file that was not yet pushed to the master branch (e.g., version Z), but upon pulling remote changes, new files with versions X and Y were added to the migration directory. In this scenario, there are two cases:

    • If the new (remote) changes do not conflict with the local changes, users can use the --exec-order non-linear flag to execute files that were added out of order (X and Y). In case of failure or unexpected behavior, the atlas schema clean command can be used to reset everything to a clean state.
    • If the new (remote) changes conflict with the local (not yet pushed) changes, the local database should be cleared using atlas schema clean, and the local changes should be adjusted to align with the pulled changes.
  • Real environment: If you encountered this issue during deployment, it means that Atlas was not set up in your CI, which is why the issue was not detected beforehand. Let's go through the steps to fix the error and set up Atlas in your CI to prevent this issue from happening again:

    • Configure Atlas to run for every PR or a change in the master branch that affects the migration directory. Learn more on how to integrate it into your GitHub Actions or GitLab CI pipelines.
    • Pull the latest changes from the master branch and run atlas migrate rebase <versions> to rebase the problematic files reported by Atlas (the "out of order" ones). Then, create a PR.
    • At this stage, if Atlas CI is set up, it will verify that the migration directory is replay-able and that the rebased migration file(s) can be executed without any issues (e.g., no conflicts with previous migrations).
    • Running atlas migrate apply (deployment) again should pass without this error.

Existing Databases

Baseline migration

If you have an existing database project and want to switch over to Atlas Versioned Migrations, you need to provide Atlas with a starting point. The first step is to create a migration file reflecting the current schema state. This can be easily done:

atlas migrate diff my_baseline \
--dir "file://migrations" \
--dev-url "docker://mysql/8/my_schema" \
--to "mysql://root:pass@localhost:3306/my_schema"

Atlas will generate a "baseline" file from the database schema. For example:

CREATE TABLE `users` (
`id` bigint(20) NOT NULL AUTO_INCREMENT,
`age` bigint(20) NOT NULL,
`name` varchar(255) COLLATE utf8mb4_bin NOT NULL,
UNIQUE KEY `age` (`age`)

Regardless of whether you added additional migration files after the baseline, you need to specify the baseline version during your first migration execution. Atlas will mark this version as already applied and proceed with the next version after it. For example:

atlas migrate apply \
--url "mysql://root:pass@localhost:3306/example" \
--dir "file://migrations" \
--baseline "20220811074144"

Allow Dirty

If your database contains resources but no revision information yet, Atlas will refuse to execute migration files. One way to override that behavior is by using the --baseline flag. However, in cases where existing tables are not managed by Atlas at all and should not be part of a baseline file, you can run the first migration execution with the --allow-dirty flag to operate on a non-clean database.

atlas migrate apply \
--url "mysql://root:pass@localhost:3306/example" \
--dir "file://migrations" \

Dry Run

If you want to check what exactly Atlas would do when attempting a migration execution, you can provide the --dry-run flag:

  • --dry-run to not execute any SQL but print it on the screen.

Down migrations

Migrations that "roll back" or reverse changes made to the database schema are called "down migrations". These are often used during local development to undo the changes made by corresponding "up migrations". Atlas follows a linear migration history model, in which all migration files are "roll-forward". However, it is still possible to clean or revert schema changes made by specific migration files using the schema clean or schema apply commands. For example:

Reverting deleted migration files

1. To undo the changes made by migration files that have already been deleted, you can use the schema apply command to roll back the database state to the state defined by the migration directory:

atlas schema apply \
--url "mysql://root:pass@localhost:3306/example" \
--to "file://migrations" \
--dev-url "docker://mysql/8/example" \
--exclude "atlas_schema_revisions"

2. Remove the files/versions that were rolled back from the revision table:

atlas migrate set \
--url "mysql://root:pass@localhost:3306/example" \
--dir "file://migrations"
Example output
Current version is 20221207170547 (2 removed):

- 20221225113035 (add_users)
- 20221225113406 (add_blog_posts)

3. Verify the migration status by using the atlas migrate status command:

atlas migrate status \
--url "mysql://root:pass@localhost:3306/example" \
--dir "file://migrations"
Example output
Migration Status: OK
-- Current Version: 20221207170547
-- Next Version: Already at latest version
-- Executed Files: 25
-- Pending Files: 0

Reverting applied migration files

1. To undo the changes made by migration files that are still present in the migration directory, you can use the version parameter to specify that the database should be rolled back to the state defined in a specific version of the migration directory. For example:

atlas schema apply \
--url "mysql://root:pass@localhost:3306/example" \
--to "file://migrations?version=20220925094437" \
--dev-url "docker://mysql/8/example" \
--exclude "atlas_schema_revisions"
tree .
└── migrations
├── 20220925092817_initial.sql
├── 20220925094021_second.sql
├── 20220925094437_third.sql
├── 20220925094438_fourth.sql # changes made by this file are reverted.
└── atlas.sum

2. To remove the 20220925094438 version from the revision table, set the current version to the preceding one 20220925094437:

atlas migrate set 20220925094437 \
--url "mysql://root:pass@localhost:3306/example" \
--dir "file://migrations"
Example output
Current version is 20220925094437 (1 removed):

- 20220925094438 (fourth)

3. Verify the migration status by using the atlas migrate status command:

atlas migrate status \
--url "mysql://root:pass@localhost:3306/example" \
--dir "file://migrations"
Example output
Migration Status: PENDING
-- Current Version: 20220925094437
-- Next Version: 20220925094438
-- Executed Files: 3
-- Pending Files: 1

Reverting all schema changes

In order to clean the entire schema, use the schema clean command, and then use migrate apply to replay the migration files. For example:

atlas schema clean -u "mysql://root:pass@:3306/example"
atlas migrate apply -u "mysql://root:pass@:3306/example"

Migration status

In addition to the --dry-run flag Atlas also provides the atlas migrate status command, that provides in-depth information about the migration status of the connected database.

Multi-Tenant environments

The Atlas configuration language provides built-in support for executing versioned workflows in multi-tenant environments. Using the for_each meta-argument, users can define a single env block that is expanded to N instances, one for each tenant:

env "prod" {
for_each = toset(var.tenants)
url = urlsetpath(var.url, each.value)
migration {
dir = "file://migrations"
format {
migrate {
apply = format(
"{{ json . | json_merge %q }}",
Tenant : each.value

Read more about how to define versioned workflows using project files in multi-tenant environments.


First time apply with baseline on production environment:

atlas migrate apply \
--env "production" \
--baseline "20220811074144"

Execute 1 pending migration file, but don't run, but print SQL statements on screen:

atlas migrate apply 1 \
--env "production" \
--baseline "20220811074144" \

Specify revision table schema and custom migration directory path:

atlas migrate apply \
--url "mysql://root:pass@remote:3306/my_database" \
--revisions-schema "atlas_migration_history" \
--dir "file://custom/path/to/dir"

Ignore unclean database and run the first 3 migrations:

atlas migrate apply 3 \
--url "mysql://root:pass@remote:3306/my_database" \
--dir "file://custom/path/to/dir"

Run all pending migrations, but do not use a transaction:

atlas migrate apply \
--url "mysql://root:pass@remote:3306/my_database" \
--tx-mode "none"

Show information about the migration status of a deployment:

atlas migrate status \
--url "mysql://root:pass@remote:3306/my_database" \
--dir "file://custom/path/to/dir" \
--revisions-schema "atlas_migration_history"


CLI Command Reference