Getting changes out of Oracle and into Kafka or a lakehouse looks done the moment the connector says RUNNING. The hard part is everything that doesn't raise an error: updates that arrive with blank columns, large LOBs that never arrive, numbers encoded as base64, and a restart position that quietly depends on archived logs a routine purge can delete. These guides treat Oracle as a source for the data platform — Debezium and LogMiner today, Kafka and Databricks downstream — and each one comes with a Docker lab that shows what lands in the topic, not just what the docs promise.
Lab One UPDATE, three tables that differ only in table-level supplemental logging: with (ALL) COLUMNS the event had the real key and every column; without it, the event said ID 0 and CUSTOMER ''. An 8,000-character CLOB update produced no event at all until lob.enabled=true. The connector was RUNNING through all of it.