What does ordering actually mean?
Kafka orders records within a partition; do not assume a total order across partitions. The same key is typically used to route related events, subject to the configured partitioner. Include the partition key, producer configuration and required order in your test contract.
Which failure cases need evidence?
Send an approved payment, a reversal and a duplicate reversal. Assert the final balance and number of side effects. Also try delayed delivery and a consumer restart. Producer idempotence does not automatically prove application-level idempotence across every downstream side effect.
What should a PoC record?
Record topic, partition count, event schema version, keys, producer settings and the precise expected state. Demonstrate how the chosen product publishes or exports events for your licensed edition; a database connector is not evidence of Kafka output support.
Candidates for this task
- DATAMIMIC Enterprise Platform: Reads a bounded set of Kafka records for transformation and can publish results to a Kafka topic. Masking is a documented use case.
- GenRocket: The vendor describes Kafka messages; verify the actual producer, configuration and target in a demonstration.
- Tonic Fabricate: Can export generated datasets; Kafka publication through a separate producer is an integration to build and test.
References and further reading
- Apache Kafka: events, keys and partitions
- Apache Kafka: producer idempotence and delivery
- GenRocket: test data and Kafka messages
- DATAMIMIC (vendor article): real-time Kafka masking for payment systems