When stopping the PG connector while it's doing a snapshot, the producer thread will not be stopped or interrupted, but keeps going.

It will eventually stop if all the records it emits fit into the internal queue. Should it emit more records than the queue can take (the usual case for a non-tiny database), the producer thread will wait indefinitely at the call to BlockingQueue#put() in PostgresConnectorTask#enqueueRecord().

Note that this also keeps the table locks acquired at the beginning of the snapshotting phase (which protect against schema changes, not data changes) until KC is shut down.