Amplify Data Migration Tool

• 日本語

Features

• Implementation of migration in TypeScript
• Execute export using DynamoDB Point-In-Time Recovery
• Management of executed migrations

Installation

npm install -D @purpom-media-lab/amplify-data-migration

Usage

Initialize

Creating a migration table

At the beginning (each app & branch), create a migration table and S3 bucket with the following command.

data-migration init --appId '<appId>' --branch '<branch name>' --profile '<profile name>'

Create Migration File

You can create a migration file template by specifying the name of the migration:

data-migration create --name <migration file name> --migrationsDir <path to migration file directory>

Update Models

Suppose the Todo model has already existed as follows.

const schema = a.schema({
  Todo: a
    .model({
      content: a.string(),
    })
    .authorization((allow) => [allow.owner()]),
});

After the release, you will need a completed field, and you need to add the completed field to the Todo model as follows.

const schema = a.schema({
  Todo: a
    .model({
      content: a.string(),
      completed: a.boolean().required(),
    })
    .authorization((allow) => [allow.owner()]),
});

The change in the model adds a completed field, but the existing data in DynamoDB table does not have the completed field. An error occurs when you try to get an existing record without the field, which is a required field, which is a required field, via AppSync.

You can implement migration processing in the implementation class with the Migration interface as follows. The following is an example of a migration that adds a completed field with the value false.

import {
  Migration,
  MigrationContext,
  ModelTransformer,
} from "@purpom-media-lab/amplify-data-migration";

export default class AddCompletedField_1725285846599 implements Migration {
  readonly name = "add_completed_field";
  readonly timestamp = 1725285846599;
  async run(context: MigrationContext) {
    type OldTodo = { id: string; content: string };
    type NewTodo = { id: string; content: string; completed: boolean };
    const transformer: ModelTransformer<OldTodo, NewTodo> = async (
      oldModel
    ) => {
      return { ...oldModel, completed: false };
    };
    await context.modelClient.updateModel("Todo", transformer);
  }
}

Put Models

Suppose you add a Profile model that did not exist before the release. The Profile model exists for each user, Since the application needs the Profile model, you need to register a Profile for each user in DynamoDB in the migration.

const schema = a.schema({
  Profile: a
    .model({
      name: a.string().requred(),
    })
    .authorization((allow) => [allow.owner()]),
});

You can use the ModelClient.putModel method to register new data in amplify-data-migration as follows The migration to register a new Profile corresponding to a user in Cognit's UserPool is as follows

import {
  Migration,
  MigrationContext,
  ModelGenerator,
} from "@purpom-media-lab/amplify-data-migration";
import {
  CognitoIdentityProviderClient,
  ListUsersCommand,
} from "@aws-sdk/client-cognito-identity-provider";

const client = new CognitoIdentityProviderClient();

type Profile = {
  id: string;
  name: string;
  owner: string;
  createdAt?: string;
  updatedAt?: string;
};

export default class AddProfileModel_1725285846601 implements Migration {
  readonly name = "add_profile_model";
  readonly timestamp = 1725285846601;
  private userPoolId: string = process.env.USER_POOL_ID!;

  async run(context: MigrationContext) {
    const userPoolId = this.userPoolId;
    const generator: ModelGenerator<Profile> = async function* () {
      let token;
      do {
        const command: ListUsersCommand = new ListUsersCommand({
          UserPoolId: userPoolId,
          Limit: 20,
          PaginationToken: token,
        });
        const response = await client.send(command);
        token = response.PaginationToken;
        for (const user of response.Users ?? []) {
          const owner = `${user.Username}::${user.Username}`;
          const now = new Date().toISOString();
          yield {
            id: crypto.randomUUID(),
            name:
              user?.Attributes?.find((attribute) => attribute.Name === "Email")
                ?.Value ?? "",
            owner,
            createdAt: now,
            updatedAt: now,
          };
        }
      } while (token);
    };
    await context.modelClient.putModel("Profile", generator);
  }
}

Run Migrations

When you run the data-migration migrate command as shown below, amplify-data-migration will run pending migrations.

data-migration migrate --appId '<appId>' --branch '<branch name>' --migrationsDir ./dist/migrations/ --profile '<profile name>'

Migrate from export data with Point-in-Time Recovery

Suppose the book model exists as follows.

const schema = a.schema({
  Book: a.model({
    author: a.string(),
    title: a.string(),
  }),
});

After the release, it is necessary to change the author, title field as a key, and suppose you have changed the Book model as follows.

const schema = a.schema({
  Book: a
    .model({
      author: a.id().required(),
      title: a.string(),
    })
    .identifier(["author", "title"])
    .authorization((allow) => [allow.owner()]),
});

If you change the model key on AWS Amplify, the DynamoDB table is replace and the existing data is deleted. Therefore, the existing data cannot be migrated by implementing the Migration.run function alone. In this case, the export of existing data is implemented with the Migration.export function.

In the Migration.export function, you can export the existing data of the model by calling context.modelClient.exportModel. And you can import the exported data into a table after replace by calling context.modelClient.runImport in Migration.run function.

import {
  ExportContext,
  Migration,
  MigrationContext,
  ModelTransformer,
} from "@purpom-media-lab/amplify-data-migration";

export default class ChangeBookKey_1725285846600 implements Migration {
  readonly name = "change_book_key";
  readonly timestamp = 1725285846600;

  async export(
    context: ExportContext
  ): Promise<Record<string, DynamoDBExportKey>> {
    // Export Book table to S3 bucket with Point-in-Time Recovery
    const key = await context.modelClient.exportModel("Book");
    return { Book: key };
  }

  async run(context: MigrationContext) {
    type OldBook = { id: string; author: string; title: string };
    type NewBook = { author: string; title: string };
    const newKeys: string[] = [];
    const transformer: ModelTransformer<OldBook, NewBook> = async (
      oldModel
    ) => {
      const { id, ...newModel } = oldModel;
      if (newKeys.includes(`${newModel.author}:${newModel.title}`)) {
        // Skip if the same key already exists.
        return null;
      }
      newKeys.push(`${newModel.author}:${newModel.title}`);
      return newModel;
    };
    // Import exported data to new Book table.
    await context.modelClient.runImport("Book", transformer);
  }
}

Run the data-migration export command as follows, and amplify-data-migration executes an export for pending migration. Usually, this command is assumed to be called before executing the deployment with npx ampx pipeline-deploy.

data-migration export --appId '<appId>' --branch '<branch name>' --profile '<profile name>'

Destroy

If you no longer want to use the Amplify Data Migration Tool, run the following command to destroy the migration table and S3 bucket.

data-migration destroy --appId '<appId>' --branch '<branch name>' --profile '<profile name>'

Development

Build

npm run build

Test

We will use LocalStack for testing. Before running the tests, please run the following command to start LocalStack.

docker-compose up

Run the test by executing the following command:

npm test