Is it possible to obtain specific message offset in Kafka+SparkStreaming?

Question

I'm trying to obtain and store the offset for a specific message in Kafka by using Spark Direct Stream. Looking at the Spark documentation is simple to obtain the range offsets for each partition but what I need is to store the start offset for each message of a topic after a full scan of the queue.

Answer 1

Yes, you can use MessageAndMetadata version of createDirectStream which allows you to access message metadata.

You can find example here which returns Dstream of tuple3.

val ssc = new StreamingContext(sparkConf, Seconds(10))

val kafkaParams = Map[String, String]("metadata.broker.list" -> (kafkaBroker))
var fromOffsets = Map[TopicAndPartition, Long]()
val topicAndPartition: TopicAndPartition = new TopicAndPartition(kafkaTopic.trim, 0)
val topicAndPartition1: TopicAndPartition = new TopicAndPartition(kafkaTopic1.trim, 0)
fromOffsets += (topicAndPartition -> inputOffset)
fromOffsets += (topicAndPartition1 -> inputOffset1)

val messagesDStream = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder, Tuple3[String, Long, String]](ssc, kafkaParams, fromOffsets, (mmd: MessageAndMetadata[String, String]) => {
    (mmd.topic ,mmd.offset, mmd.message().toString)
  })

In above example tuple3._1 will have topic, tuple3._2 will have offset and tuple3._3 will have message.

Hope this helps!