Finally I found how to configure the agent in order to assign the key for partitionin in kafka topic using the absolute path of the files in the data source. I post the answer in case someone will need it. It is necessary to set the property 'fileHeaderKey=key". In this way, when the event is passed to kafka sink the header contains the pairs key =absolute/path/of /the/file and kafka can use it as the key in its message. agent3a.sources= source3a agent3a.channels= channel3a agent3a.sinkss= sink3a source agent3a.sources.source3a.type = TAILDIR agent3a.sources.source3a.filegroups= f1 agent3a.sources.source3a.filegroups.f1 = /path/local/data/source/ agent3a.sources.source3a.channels= channel3a agent3a.sources.source3a.batchSize=1 agent3a.sources.source3a.fileHeader= True agent3a.sources.source3a.fileHeaderKey= key #####property to set the fileHeader the key for partitioning### Memory Channel agent3a.channels.channel3a.type = memory agent3a.channels.channel3a.transactionCapacity = 100 agent3a.channels.channel3a.capacity = 100 Sink agent3a.sinks.sink3a.type= org.apache.flume.sink.kafka.KafkaSink agent3a.sinks.sink3a.kafka.bootstrap.servers =########hostname1#####:port, ########hostname2#####:port agent3a.sinks.sink3a.kafka.topic= topic_test agent3a.sinks.sink3a.kafka.producer.acks=1 agent3a.sinks.sink3a.channel= channel3a agent3a.sinks.sink3a.FlumeBatchSize=1 agent3a.sinks.sink3a.useFlumeEventFormat = true agent3a.sinks.sink3a.kafka.producer.batch.size=10  It is, alternatively possibile to set the key for partitioning also using interceptors: it is only necessary to set its property '.key=key' bye 🙂 ... View more