Debezium

Use it when

• You need CDC from many databases under one open-source roof: MySQL, PostgreSQL, MongoDB, SQL Server, Oracle, Db2, and Cassandra through one connector model.
• Kafka is already the backbone; Debezium deploys as Connect source connectors and lands ordered change events in topics.
• Downstream patterns vary: warehouse feeds, cache and index invalidation, and the outbox pattern for service integration are all established uses.
• You want the community default; most CDC platforms here either embed it or compete with it.

Think twice when

• Nobody wants to run Kafka Connect; a Connect cluster is real operational surface (Debezium Server and the embedded engine avoid it at reduced scope, and Maxwell or Sequin are lighter for single-source cases).
• Exactly-once is assumed; delivery is at-least-once, and consumers must be idempotent.
• You want managed CDC with support; Debezium arrives via a vendor (Confluent, Red Hat, cloud connectors) rather than from the project.

How it runs

Connectors read transaction logs (Postgres WAL, MySQL binlog, and equivalents) and publish change events through a Kafka Connect cluster; incremental snapshots backfill via signals, and a history topic records source DDL. Debezium Server targets Kinesis, Pub/Sub, Pulsar, and Redis without Kafka.