Architecture

KSI is a filter created to run within the Kroxylicious Kafka proxy (https://github.com/kroxylicious/kroxylicious). The filter utilises an embedded or external Iceberg engine to fetch data before reformatting it for Kafka purposes.

A typical flow may look like this:

1. Kroxylicious receives a FETCH request for offsets that are stored in Iceberg not Kafka

2. This satisfies the routing config in Kroxylicious and so the request is intercepted by the K.S.I. filter

3. In parallel the filter:

   1. Fetches data matching the request from Iceberg using a provided Iceberg engine

   2. Fetches the appropriate schema of the data of the data from Schema Registry

   3. Launches a pre-fetch task to fetch the next chunk of data from Iceberg (K.S.I. expects sequential access by Kafka clients and can greatly increase performance by fetching data to satisfy future requests ahead of time).

4. K.S.I. encodes the Iceberg data into Kafka message format

5. The K.S.I. filter passes the encoded response back to Korxylicious to return to the client.