Connectors

Kafka is the only connector StreamFusion considers production-grade today, across every format Flink itself supports for it. It's covered in depth in this section.

Parquet (a local file source, and a sink to any filesystem Flink supports) and Fluss (log-table source ingest) also have native paths, but neither has seen the same production hardening as Kafka — treat them as experimental.

Formats

A format decodes or encodes the bytes on the wire; it's independent of which connector carries them (though not every connector supports every format — see each connector's page). StreamFusion ships a native implementation for:

• JSON
• CSV
• Avro
• Avro Confluent (schema-registry-backed Avro)
• Protobuf
• Raw (the single-column passthrough format)
• CDC JSON (Debezium, OGG, Maxwell, and Canal's changelog envelopes)

Each optional format is its own streamfusion-* Maven artifact and native library, mirroring Flink's own connector/format module split — installing a format you don't use never pulls its native code into a connector you do. A missing format module is always a normal planner fallback to Flink's own decode/encode path, never a linkage failure. See Deployment for which JARs a given connector+format combination needs.