Spark2.x 自从引入了 Structured Streaming 后，未来数据操作将逐步转化到 DataFrame/DataSet，以下将介绍 Spark2.x 如何与 Kafka0.10+整合

Structured Streaming + Kafka

1. 引包

groupId = org.apache.spark
artifactId = spark-sql-kafka-0-10_2.11
version = 2.1.1

为了让更直观的展示包的依赖，以下是我的工程 sbt 文件

name := "spark-test"
version := "1.0"
scalaVersion := "2.11.7"
// https://mvnrepository.com/artifact/org.apache.spark/spark-core_2.11
libraryDependencies += "org.apache.spark" % "spark-core_2.11" % "2.1.1" % "provided"
// https://mvnrepository.com/artifact/org.apache.spark/spark-mllib_2.11
libraryDependencies += "org.apache.spark" % "spark-mllib_2.11" % "2.1.1" % "provided"
// https://mvnrepository.com/artifact/org.apache.spark/spark-streaming_2.11
libraryDependencies += "org.apache.spark" % "spark-streaming_2.11" % "2.1.1" % "provided"
// https://mvnrepository.com/artifact/org.apache.hadoop/hadoop-client
libraryDependencies += "org.apache.hadoop" % "hadoop-client" % "2.7.3"
// https://mvnrepository.com/artifact/mysql/mysql-connector-java
libraryDependencies += "mysql" % "mysql-connector-java" % "5.1.38"
// https://mvnrepository.com/artifact/org.apache.kafka/kafka_2.11
libraryDependencies += "org.apache.kafka" % "kafka_2.11" % "0.10.2.1"
//libraryDependencies += "org.apache.spark" % "spark-streaming-kafka-0-10_2.11" % "2.1.1"
libraryDependencies += "org.apache.spark" % "spark-sql-kafka-0-10_2.11" % "2.1.1"

1. Structured Streaming 连接 Kafka

def main(args: Array[String]): Unit = {

    val spark = SparkSession
      .builder()
      .appName("Spark structured streaming Kafka example")
      //      .master("local[2]")
      .getOrCreate()

    val inputstream = spark.readStream
      .format("kafka")
      .option("kafka.bootstrap.servers", "127.0.0.1:9092")
      .option("subscribe", "testss")
      .load()
    import spark.implicits._
    val query = inputstream.select($"key", $"value")
      .as[(String, String)].map(kv => kv._1 + " " + kv._2).as[String]
      .writeStream
      .outputMode("append")
      .format("console")
      .start()

    query.awaitTermination()
  }

流的元数据如下

可配参数

不能配置的参数

• group.id: 对每个查询，kafka 自动创建一个唯一的 group

• auto.offset.reset: 可以通过 startingOffsets 指定，Structured Streaming 会对任何流数据维护 offset, 以保证承诺的 exactly once.

• key.deserializer: 在 DataFrame 上指定，默认 ByteArrayDeserializer

• value.deserializer: 在 DataFrame 上指定，默认 ByteArrayDeserializer

• enable.auto.commit:

• interceptor.classes:

Stream + Kafka

1. 从最新offset开始消费

   def main(args: Array[String]): Unit = {
   	val kafkaParams = Map[String, Object](
   	  "bootstrap.servers" -> "localhost:9092",
   	  "key.deserializer" -> classOf[StringDeserializer],
   	  "value.deserializer" -> classOf[StringDeserializer],
   	  "group.id" -> "use_a_separate_group_id_for_each_stream",
   	  "auto.offset.reset" -> "latest",
   	  "enable.auto.commit" -> (false: java.lang.Boolean)
   	)
   
   	val ssc =new StreamingContext(OpContext.sc, Seconds(2))
   	val topics = Array("test")
   	val stream = KafkaUtils.createDirectStream[String, String](
   	  ssc,
   	  PreferConsistent,
   	  Subscribe[String, String](topics, kafkaParams)
   	)
   	stream.foreachRDD(rdd=>{
   	  val offsetRanges=rdd.asInstanceOf[HasOffsetRanges].offsetRanges
   	  rdd.foreachPartition(iter=>{
   		val o: OffsetRange = offsetRanges(TaskContext.get.partitionId)
   		println(s"${o.topic} ${o.partition} ${o.fromOffset} ${o.untilOffset}")
   	  })
   	  stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges)
   	})
   
   //    stream.map(record => (record.key, record.value)).print(1)
   	ssc.start()
   	ssc.awaitTermination()
     }

2. 从指定的offset开始消费

   def main(args: Array[String]): Unit = {
     val kafkaParams = Map[String, Object](
   	"bootstrap.servers" -> "localhost:9092",
   	"key.deserializer" -> classOf[StringDeserializer],
   	"value.deserializer" -> classOf[StringDeserializer],
   	"group.id" -> "use_a_separate_group_id_for_each_stream",
   	//      "auto.offset.reset" -> "latest",
   	"enable.auto.commit" -> (false: java.lang.Boolean)
     )
     val ssc = new StreamingContext(OpContext.sc, Seconds(2))
     val fromOffsets = Map(new TopicPartition("test", 0) -> 1100449855L)
     val stream = KafkaUtils.createDirectStream[String, String](
   	ssc,
   	PreferConsistent,
   	Assign[String, String](fromOffsets.keys.toList, kafkaParams, fromOffsets)
     )
   
     stream.foreachRDD(rdd => {
   	val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges
   	for (o <- offsetRanges) {
   	  println(s"${o.topic} ${o.partition} ${o.fromOffset} ${o.untilOffset}")
   	}
   	stream.asInstanceOf[CanCommitOffsets].commitAsync(offsetRanges)
     })
   
     //    stream.map(record => (record.key, record.value)).print(1)
     ssc.start()
     ssc.awaitTermination()
   }