Ddb Migration Tools

ddb-migration-tools

Use ddb-migration-tools for TypeScript DynamoDB schema/data migrations that need stage-aware configuration, checksum drift detection, resumable checkpoints, and a shared per-account/per-region ledger table.

Start Here

1. Inspect the target project before editing:
   • package.json
   • existing ddb-migrations.config.*
   • migrations/
   • deployment docs or stack definitions for DynamoDB table names
2. Install the library if missing:

   npm install --save-dev ddb-migration-tools
   

3. Initialize only when no config exists:

   npx ddb-migrate init
   

4. Prefer directory migrations for non-trivial work:

   npx ddb-migrate create "seed counter metadata"
   

Configuration Pattern

Use one ledger table per AWS account and region. Do not put accountId in the ledger primary key; the account is implied by the table being written to.

Recommended shape:

{
  "appName": "my-app",
  "migrationsDir": "migrations",
  "ledger": {
    "tableName": "ddb-migrations-ledger"
  },
  "stages": {
    "dev": {
      "region": "us-east-1",
      "tablePrefix": "my-app-dev-"
    },
    "staging": {
      "region": "us-east-1",
      "accountId": "123456789012",
      "tablePrefix": "my-app-staging-"
    },
    "prod": {
      "region": "us-east-1",
      "accountId": "210987654321",
      "tablePrefix": "my-app-prod-"
    }
  }
}

Use tables overrides when table names do not share a prefix. Use ledger.scope only when multiple apps intentionally share a ledger namespace.

If app tables span multiple regions but the project wants one shared ledger, set ledger.region to where the ledger table lives. Each stage still talks to its app tables in stage.region; ledger reads/writes route to ledger.region. Per-stage ledgerRegion / ledgerEndpoint override the global value. Only reach for this when stages actually use different regions — for single-region projects the default (ledger follows stage) is correct.

Writing Migrations

Import the context type from the package:

import type { MigrationContext } from 'ddb-migration-tools';

export const description = 'Backfill schemaVersion on User items';

export async function up(ctx: MigrationContext): Promise<void> {
  const Users = ctx.tableName('users');
  ctx.logger.info(`Using ${Users}`);
}

Use ctx.ddb for item operations and ctx.ddbRaw for table-level operations such as GSI changes. Resolve all table names through ctx.tableName(logical).

For long migrations, ctx.sdkStats.snapshot() reports observed app-table DynamoDB send() calls, reads/writes, pages, returned items, failures, throttles, and optional consumed capacity. These are top-level SDK calls, not internal retry attempts, and do not include ledger/checkpoint writes.

Honor dry-run explicitly:

if (ctx.dryRun) {
  ctx.logger.info('would update item X');
  return;
}

For long scans or backfills, use checkpoints:

const checkpoint = await ctx.getCheckpoint<{ lastKey?: Record<string, unknown> }>();
await ctx.checkpoint({ lastKey: page.LastEvaluatedKey });

Safety Rules

• Make migrations idempotent. Use conditional writes or filters so reruns do not corrupt data.
• Treat down() as optional. Throw down() not implemented when rollback would be unsafe or fake.
• Never edit, reorder, or rename a migration after it has been applied to any shared environment; checksum drift should block that.
• Prefer forward-only repair migrations over mutating old migration files.
• Avoid DynamoDB scans unless the migration is explicitly a controlled backfill. For scans, page results, checkpoint, and consider segmented scans.
• Do not print secrets or full customer records in migration logs.
• Check production credentials/profile selection before running up --stage prod.

Runbook

Before applying:

npx ddb-migrate status --stage staging
npx ddb-migrate up --stage staging --dry-run

Apply:

npx ddb-migrate up --stage staging
npx ddb-migrate status --stage staging

Use SDK observability flags when diagnosing long backfills:

npx ddb-migrate up --stage staging --capacity      # request consumed capacity in responses
npx ddb-migrate up --stage staging --no-sdk-stats  # disable SDK stats for this run

Rollback only when down() is intentionally implemented and safe:

npx ddb-migrate down --stage staging --shift 1 --dry-run
npx ddb-migrate down --stage staging --shift 1

Use --to <migrationId> to stop at a specific migration during staged rollout. Use --shift N to roll back the newest N completed migrations.

Ledger Stack

If the project needs a ledger table, use the included stack/ reference stack from this repo. Deploy one stack per account/region with Serverless Framework. Prefer osls when the project standard requires it.

Do not remove ledger stacks unless the user explicitly asks. The reference stack retains the DynamoDB table on stack deletion, but removal is still an operational action.