Databricks Zerobus is a new serverless streaming ingestion mechanism that lets applications push event data directly into Delta Lake tables via gRPC and Apache Arrow Flight, bypassing the need for Kafka clusters, connectors, and traditional streaming infrastructure. A hands-on exploration covers setup requirements (workspace, Delta table, OAuth auth, SDK), a practical test streaming ~140K records from the Divvy bike dataset, and real friction points encountered: Serverless compute couldn't reach the Zerobus endpoint, and schema compatibility required iteration. Data landed successfully after switching to an all-purpose cluster. A key operational concern flagged is small-file proliferation inherent to streaming workloads — OPTIMIZE/compaction strategies remain necessary. The verdict: Zerobus meaningfully simplifies lakehouse streaming for organizations that don't need full Kafka capabilities, though questions around scale, performance, and production best practices remain open.
