This needs double-checking, but I can't see how any exceptions on the producer side of the MongoDB connector would be propagated to the polling loop by Kafka Connect. I.e. it seems as if the connector would keep running in case of a producer exception.