shortishly

You could use the proxy to spread topics over a number of brokers. The broker and proxy share a number of services and layers, that could be used to route:

https://blog.tansu.io/articles/route-layer-service

My itch for SQLite was smaller scale (and reproducible) environments, e.g., development, test/integration (with a single file to reset the environment). PostgreSQL was intended for "larger scale", with (database level) partitioning of Kafka records on each topic/partition, and replication for leader/follower setups, which might work better for HA. S3 for environments where latency is less of any issue (though with the SlateDB/S3 engine that might change).

S3: Not yet. I've been working through tuning each engine, S3 is next on the list.

I agree that it isn't straight forward! Tansu uses the JSON protocol descriptors from Apache Kafka, generating ~60k LoC of Rust to represent the structures. It then uses a custom Serde encoder/decoder to implement the protocol: original, flexible and tag buffers formats for every API version (e.g., the 18 just in FETCH). It is based off spending the past ~10 years using Kafka, and writing/maintaining an Erlang client (there are no "good" Kafka clients for Erlang!). It also uses a bunch of collected protocol examples, to encode/decode during the tests. Tansu is also a Kafka proxy, which is also used to feed some of those tests.

Some of the detail: https://blog.tansu.io/articles/serde-kafka-protocol

However, there are definitely cases I am sure where Tansu isn't compatible. For example, Kafka UI (kafbat) reports a strange error when doing a fetch (despite actually showing the fetched data), which I've yet to get to the bottom of.

If you find any compatibility issues, then please raise an issue, and I can take a look.

Yes: with a schema backed topic (AVRO, JSON or Protocol buffer) Tansu can write to Apache Iceberg, Delta or Parquet. You can use a Sink topic to write directly to an open table format (including Parquet) skipping (most of) the Kafka metadata.

https://blog.tansu.io/articles/parquet

Tansu author here. Storage is a pluggable choice of: PostgreSQL, memory, SQLite or S3. There are others in the pipeline (SlateDB, ...).