การแปลงข้อความเดี่ยว (SMTs)

Single Message Transforms (SMTs) are lightweight functions applied to each record as it flows through a Kafka Connect connector.

They let you tweak data — rename fields, add metadata, route topics — without a separate processing job.

SMTs run inside the connector worker:

• For a source connector, after reading from the system and before writing to Kafka.
• For a sink connector, after reading from Kafka and before writing to the target.

Transforms are configured declaratively in the connector's JSON or properties. You name the chain, then configure each transform.

Kafka Connect ships many useful transforms:

• InsertField — add a static or metadata field.
• ReplaceField — rename or drop fields.
• MaskField — redact sensitive values.
• ExtractField — pull one field up as the whole value.

RegexRouter rewrites the destination topic name, handy for adding prefixes when ingesting many tables.

List multiple transforms separated by commas; they run in order, each receiving the previous one's output.

Many SMTs come in two forms, suffixed with $Key or $Value, so you can target the record's key or value independently.

For example InsertField$Key versus InsertField$Value.

Predicates let a transform apply conditionally — for example, only to tombstone records or to a specific topic.

SMTs operate on one record at a time. They cannot:

• Join across records or topics.
• Aggregate or window.
• Call external services efficiently.

For those, use Kafka Streams or ksqlDB instead.

You can implement your own by writing a class that implements the Transformation interface, packaging it as a JAR, and placing it on the worker plugin path.

Use SMTs wisely:

• Keep chains short and readable.
• Mask sensitive data as early as possible.
• Prefer predicates over per-topic connectors when filtering.
• Move heavy logic to a stream processor.

You learned about Single Message Transforms.

• SMTs reshape each record inline within a connector.
• Use built-ins like InsertField, ReplaceField, MaskField, RegexRouter.
• Chain them and gate with predicates.
• For joins, aggregations, or windows, use a stream processor instead.