This article explores Kafka's transition toward a cloud-native architecture, examining how tiered storage, FinOps telemetry, elastic consumer scaling, virtual clusters, and Share Groups reshape the operational and economic model of event streaming platforms. It also analyzes emerging diskless-storage proposals and their architectural trade-offs.

Schema proliferation builds slowly and gets expensive fast. One schema per event type feels right until there are ten tables, union queries spanning all of them, and a single field rename touching every schema. Discriminator-based schema consolidation collapses that to two tables, turning multi-table unions into a single query, while new variants are additive and don't break existing consumers.