AWS Database Migration Service

Use it when

• The job is getting a database into AWS, with replication keeping the target current until cutover; migration is what DMS is built for.
• Sources and targets are heterogeneous; relational, warehouse, and NoSQL engines on either side, with Schema Conversion handling the translation.
• You want AWS to run the replication layer, sized as an instance or consumed serverless as capacity units.
• Ongoing CDC into S3, Kinesis, or an AWS database is useful after the migration completes.

Think twice when

• CDC is the product, not the migration; purpose-built tools (Debezium for open source, Datastream on Google Cloud) handle schema drift and streaming semantics more sharply, and DMS backfills no altered columns.
• Tables lack primary keys; delivery is at-least-once and duplicates surface without them.
• Latency budgets are tight and sustained; a replication instance is capacity you size and watch.

How it runs

A managed replication instance (or DMS Serverless capacity units) runs tasks between a source and a target endpoint, full load first, CDC after, with automatic failover to a standby. Sources need log access: ROW binlog on MySQL, supplemental logging on Oracle, MS-CDC on SQL Server. Billing is instance hours or DCU-hours plus storage.