Kafka+SparkStreaming已经发展为一个比较成熟的实时日志收集与计算架构，利用Kafka，即可以支持将用于离线分析的数据流到HDFS，又可以同时支撑多个消费者实时消费数据，包括SparkStreaming。然而，在SparkStreaming程序中如果有复杂业务逻辑的统计，使用scala代码实现起来比较困难，也不易于别人理解。但如果在SparkSteaming中也使用SQL来做统计分析，是不是就简单的多呢？

本文介绍将SparkSQL与SparkStreaming结合起来，使用SQL完成实时的日志数据统计。SparkStreaming程序以yarn-cluster模式运行在YARN上，不单独部署Spark集群。

环境部署
Hadoop-2.6.0-cdh5.8.0（YARN）
spark-2.1.0-bin-hadoop2.6
kafka-0.10.2+kafka2.2.0

实时统计需求
以10秒为间隔，统计10秒内的各大区潜客的数量
pom

 <dependency>
            <groupId>org.apache.spark</groupId>
            <artifactId>spark-streaming_${spark.artifact}</artifactId>
            <version>${spark.version}</version>
            <scope>${dependency.scope}</scope>
        </dependency>
        <dependency>
            <groupId>org.apache.spark</groupId>
            <artifactId>spark-streaming-kafka-0-10_2.11</artifactId>
            <version>${spark.version}</version>
            <scope>${dependency.scope}</scope>
        </dependency>
        <dependency>
            <groupId>com.oracle</groupId>
            <artifactId>ojdbc6</artifactId>
            <version>11.2.0.3</version>
            <scope>${dependency.scope}</scope>
        </dependency>

SparkStreaming程序代码

package com.chumi.dac.sp.stream.sparksqlcount

import com.chumi.dac.sp.stream.jdbc.DBCustomerStream
import com.chumi.dac.sp.stream.utils.DateUtil
import org.apache.kafka.clients.consumer.ConsumerRecord
import org.apache.spark.streaming._
import org.apache.spark.streaming.kafka010.{ConsumerStrategies, KafkaUtils, LocationStrategies}
import org.apache.spark.{ SparkConf, SparkContext}
import org.apache.kafka.common.serialization.StringDeserializer
import org.apache.spark.broadcast.Broadcast
import org.apache.spark.sql.{DataFrame, SQLContext}
import org.apache.spark.streaming.dstream. InputDStream

/**
  * Created by LHX on 2018/8/24 14:37.
  */

object CustomerStreamRsscCount {
    /**
      * BroadcastWrapper,用来注册广播变量。
      */
    object BroadcastWrapper {
        @volatile private var instance:Broadcast[String]=null
        def getInstance(sc: SparkContext): Broadcast[String] = {
            val point_time: String = DateUtil.getPointTime()
            if (instance == null) {
                synchronized {
                    if (instance == null) {
                        instance = sc.broadcast(point_time)
                        println("==初始化全局变量=="+point_time)
                    }
                }
            }
            instance
        }
        def update(sc: SparkContext, blocking: Boolean = false,hv:String): Broadcast[String] = {
            if (instance != null)
                instance.unpersist(blocking)
            instance = sc.broadcast(hv)
            println("==更新=="+hv)
            instance
        }
    }
    /**
      * SQLContextSingleton
      */
    object SQLContextSingleton {
        @transient  private var instance: SQLContext = _
        def getInstance(sparkContext: SparkContext): SQLContext = {
            if (instance == null) {
                instance = new SQLContext(sparkContext)
            }
            instance
        }
    }
    case class DapLog(CIM_ID:String, ENTITY_CODE:String, CARD_FOUND_TIME:String)

    def main(args: Array[String]) {

        def functionToCreateContext(): StreamingContext = {
            val conf = new SparkConf().setAppName("CustomerStreamRsscCount").setMaster("local[2]")
            val ssc = new StreamingContext(conf, Seconds(10))
            val sqlContext = SQLContextSingleton.getInstance(ssc.sparkContext)

            //要使用updateStateByKey方法，必须设置Checkpoint。
            ssc.checkpoint("C:/tmp/checkPointPath")

            //TM_SST
            val jdbcMaps = Map("url" -> "jdbc:oracle:thin:@//IP:1521/test",
                "user" -> "user",
                "password" -> "password",
                "dbtable" -> "TM_SST",
                "driver" -> "oracle.jdbc.driver.OracleDriver")
            val jdbcDFs = sqlContext.read.options(jdbcMaps).format("jdbc").load
            jdbcDFs.createOrReplaceTempView("TM_SST")

            //TM_RSSC
            val jdbcMapc = Map("url" -> "jdbc:oracle:thin:@//IP:1521/test",
                "user" -> "user",
                "password" -> "password",
                "dbtable" -> "TM_RSSC",
                "driver" -> "oracle.jdbc.driver.OracleDriver")
            val jdbcDFv = sqlContext.read.options(jdbcMapc).format("jdbc").load
            jdbcDFv.createOrReplaceTempView("TM_RSSC")

            val topics = "topic1" //stream_test01 topic1
            val topicsSet = topics.split(",").toSet
            val brokers = "IP:9095"

            val kafkaParams = Map[String, Object]("bootstrap.servers" -> brokers
                , "auto.offset.reset" -> "latest"
                , "sasl.kerberos.service.name" -> "kafka"
                , "key.deserializer" -> classOf[StringDeserializer]
                , "value.deserializer" -> classOf[StringDeserializer]
                , "group.id" -> "testgroup"
            )


            val dStream: InputDStream[ConsumerRecord[String, String]] = KafkaUtils.createDirectStream[String, String](ssc,
                LocationStrategies.PreferConsistent,
                ConsumerStrategies.Subscribe[String, String](topicsSet, kafkaParams))
            val value = dStream.transform(rdd => {
                val sqlC = SQLContextSingleton.getInstance(rdd.sparkContext)
                import sqlContext.implicits._
                val logDataFrame = rdd.map(w => {
                    val m: Array[String] = w.value().split(",")
                    DapLog(m(0), m(1), m(9))
                }).toDF()
                // 注册为tempTable
                logDataFrame.createOrReplaceTempView("TT_CUSTOMER")
                val sql = "select R.RSSC_ID,R.RSSC_NAME,COUNT(1) FROM TT_CUSTOMER  T join  TM_SST S on T.ENTITY_CODE = S.ENTITYCODE join TM_RSSC R ON S.RSSC_ID = R.RSSC_ID  GROUP BY R.RSSC_ID,R.RSSC_NAME"
                val data1: DataFrame = sqlC.sql(sql)
                val a =data1.rdd.map{r =>(r(1).toString,r(2).toString.toInt) }
                a
            })
            //将以前的数据和最新10s的数据进行求和
            val addFunction = (currValues : Seq[Int],preVauleState : Option[Int]) => {
                val currentSum = currValues.sum
                val previousSum = preVauleState.getOrElse(0)
                Some(currentSum + previousSum)
            }
            val total = value.updateStateByKey[Int](addFunction)
            //输出总计的结果
            total.print()
            ssc
    }
    //重启streamingContext,读取以前保存的数据，否则创建新的StreamingContext
    val context = StreamingContext.getOrCreate("checkPoint", functionToCreateContext _)
    context.start()
    context.awaitTermination()

    }
}

总结
其中广播变量是后期根据时间筛选时候使用的，整体思路是先读取oracle数据并注册成临时表，后获取kafka数据，根据dStream.transform()方法把数据转换成想要的结果，最后用updateStateByKey()方法累加上一批次的统计结果。 对于初学者很多sparkstream方法还不是很熟悉，所以写代码想不到使用，如果对大家有所帮助，记得点赞哦~