Full automatic dynamic lakehouse sink

[hpvd]

Flink just released a full dynamic sink for iceberg which makes writing Kafka to iceberg at scale really easy:

• automated schema evolution
• automated new topic recovery
• routing multiple types from one topic
  ...

see https://flink.apache.org/2025/11/11/from-stream-to-lakehouse-kafka-ingestion-with-the-flink-dynamic-iceberg-sink/

Are there plans to make something like this possible in combination with all the Fluss advantages?

Looks very adventurous to me especially in combination with lance format / lancedb, which easily allow evolution without data rewrite, fast point query and dealing with AI data of any modality...

[polyzos]

@MehulBatra @luoyuxia, maybe you can help with this?

[wuchong]

Hi @hpvd,

The current lakehouse integration in Fluss is indeed "dynamic", as you expected:

• Automated schema evolution: Any schema changes to a Fluss table are propagated to the corresponding Iceberg table in real time.
• Automated new table tiering: Any new Fluss table with table.datalake.enabled=true is automatically tiered to an Iceberg table by the Tiering Service (you don't need to start new jobs for the Tiering Service).
• Routing multiple types from a single topic: This is not supported in Fluss, as Fluss follows a schema-centric table model, each table has a well-defined schema, and Fluss does not mix records of different tables within one table.