A stream represents events; a table represents the current value for each key. An inner stream-table join enriches an arriving event when a matching table value exists at processing time. A missing key produces no inner-join output. A later table update does not automatically replay all earlier stream events in this simple model. Real topologies also need compatible serialization, partitioning and recovery settings. These tasks join a fictional event stream to a small current-state table and preserve only the matched record. Inspect the enriched value as well as the output count. A plain count or filter cannot substitute for evidence that the lookup contributed its current value.
Inspect the topics, partition logs and group offsets before changing the cluster.