Kafka connector

Use the kafka: scheme. Kafka is sink-only and the maintained capability is complete-mode publishing. Rows are serialized as JSON messages; topic, partition, key, and producer settings must match the consumer contract.

Set the required topic with sink.connect.parameter.topic. An optional sink.connect.parameter.partition routes every record to one numeric partition, and sink.connect.parameter.key supplies one fixed producer key. Without an explicit partition, Kafka’s producer decides placement. Without a key, ordering is only the ordering provided by the selected partition and producer configuration.

ReplicaDB serializes each row as a JSON object using sink column names. A single column named json is parsed as an existing JSON object; other rows are converted by JDBC type, with UTC timestamp text and binary content encoded for JSON. Validate the resulting schema with the consumer before a large publish.

Use environment-managed client properties for TLS and authentication. Kafka is not a table sink and does not provide complete-atomic staging; design replay, retention, and downstream idempotency around the topic contract. A failed run does not retract messages already acknowledged by the broker.