Currently the kafka input operators support a setting to restrict number of tuples per window for throttling the input in case downstream operators cannot handle the input bursts. In certain application scenarios the kafka message sizes are not similar and could vary significantly from small to large. In those case a setting based on message sizes is more appropriate like a maximum total message size per window.