大家好，又见面了，我是你们的朋友全栈君。

多数据源Join思路

多数据源Join大致有以下三种思路：

• 数据源端Join，如Android/IOS客户端在上报用户行为数据时就获取并带上用户基础信息。

• 计算引擎上Join,如用Spark Streaming、Flink做Join。

• 结果端Join，如用HBase/ES做Join,Join键做Rowkey/_id，各字段分别写入列簇、列或field。

三种思路各有优劣，使用时注意一下。这里总结在计算引擎Spark Streaming上做Join。

Stream-Static Join

流与完全静态数据Join

流与完全静态数据Join。有两种方式，一种是RDD Join方式，另一种是Broadcast Join(也叫Map-Side Join)方式。

RDD Join 方式

思路：RDD Join RDD 。

package com.bigData.spark

import com.alibaba.fastjson.{ 
   JSON, JSONException, JSONObject}
import org.apache.kafka.common.serialization.StringDeserializer
import org.apache.log4j.{ 
   Level, Logger}
import org.apache.spark.SparkConf
import org.apache.spark.streaming.kafka010.{ 
   ConsumerStrategies, KafkaUtils, LocationStrategies}
import org.apache.spark.streaming.{ 
   Durations, StreamingContext}

/** * Author: Wang Pei * License: Copyright(c) Pei.Wang * Summary: * * Stream-Static Join * * spark 2.2.2 * */
case class UserInfo(userID:String,userName:String,userAddress:String)
object StreamStaicJoin { 
   
  def main(args: Array[String]): Unit = { 
   

    //设置日志等级
    Logger.getLogger("org").setLevel(Level.WARN)

    //Kafka 参数
    val kafkaParams= Map[String, Object](
      "bootstrap.servers" -> "localhost:9092",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[StringDeserializer],
      "auto.offset.reset" -> "latest",
      "enable.auto.commit" -> (true: java.lang.Boolean),
      "group.id" -> "testTopic3_consumer_v1")

    //spark环境
    val sparkConf = new SparkConf().setAppName(this.getClass.getSimpleName.replace("$","")).setMaster("local[3]")
    val ssc = new StreamingContext(sparkConf,Durations.seconds(10))

    /** 1) 静态数据: 用户基础信息*/
    val userInfo=ssc.sparkContext.parallelize(Array(
      UserInfo("user_1","name_1","address_1"),
      UserInfo("user_2","name_2","address_2"),
      UserInfo("user_3","name_3","address_3"),
      UserInfo("user_4","name_4","address_4"),
      UserInfo("user_5","name_5","address_5")
    )).map(item=>(item.userID,item))


    /** 2) 流式数据: 用户发的tweet数据*/
    /** 数据示例： * eventTime：事件时间、retweetCount：转推数、language：语言、userID：用户ID、favoriteCount：点赞数、id：事件ID * {"eventTime": "2018-11-05 10:04:00", "retweetCount": 1, "language": "chinese", "userID": "user_1", "favoriteCount": 1, "id": 4909846540155641457} */

    val kafkaDStream=KafkaUtils.createDirectStream[String,String](
      ssc,
      LocationStrategies.PreferConsistent,
      ConsumerStrategies.Subscribe[String,String](Set("testTopic3"),kafkaParams)
    ).map(item=>parseJson(item.value())).map(item=>{ 
   
      val userID = item.getString("userID")
      val eventTime = item.getString("eventTime")
      val language= item.getString("language")
      val favoriteCount = item.getInteger("favoriteCount")
      val retweetCount = item.getInteger("retweetCount")
      (userID,(userID,eventTime,language,favoriteCount,retweetCount))
    })


    /** 3) 流与静态数据做Join (RDD Join 方式)*/
    kafkaDStream.foreachRDD(_.join(userInfo).foreach(println))

    ssc.start()
    ssc.awaitTermination()

  }

  /**json解析*/
  def parseJson(log:String):JSONObject={ 
   
    var ret:JSONObject=null
    try{ 
   
      ret=JSON.parseObject(log)
    }catch { 
   
      //异常json数据处理
      case e:JSONException => println(log)
    }
    ret
  }

}

Broadcast Join 方式

思路：RDD遍历每一条数据，去匹配广播变量中的值。

package com.bigData.spark

import com.alibaba.fastjson.{ 
   JSON, JSONException, JSONObject}
import org.apache.kafka.common.serialization.StringDeserializer
import org.apache.log4j.{ 
   Level, Logger}
import org.apache.spark.{ 
   SparkConf, SparkContext}
import org.apache.spark.streaming.kafka010.{ 
   ConsumerStrategies, KafkaUtils, LocationStrategies}
import org.apache.spark.streaming.{ 
   Durations, StreamingContext}

/** * Author: Wang Pei * License: Copyright(c) Pei.Wang * Summary: * * Stream-Static Join * * spark 2.2.2 * */
case class UserInfo(userID:String,userName:String,userAddress:String)
object StreamStaticJoin2 { 
   
  def main(args: Array[String]): Unit = { 
   

    //设置日志等级
    Logger.getLogger("org").setLevel(Level.WARN)

    //Kafka 参数
    val kafkaParams= Map[String, Object](
      "bootstrap.servers" -> "localhost:9092",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[StringDeserializer],
      "auto.offset.reset" -> "latest",
      "enable.auto.commit" -> (true: java.lang.Boolean),
      "group.id" -> "testTopic3_consumer_v1")

    //spark环境
    val sparkConf = new SparkConf().setAppName(this.getClass.getSimpleName.replace("$","")).setMaster("local[3]")
    val ssc = new StreamingContext(sparkConf,Durations.seconds(10))

    /** 1) 静态数据: 用户基础信息。 将用户基础信息广播出去。*/
    val broadcastUserInfo=ssc.sparkContext.broadcast(
      Map(
        "user_1"->UserInfo("user_1","name_1","address_1"),
        "user_2"->UserInfo("user_2","name_2","address_2"),
        "user_3"->UserInfo("user_3","name_3","address_3"),
        "user_4"->UserInfo("user_4","name_4","address_4"),
        "user_5"->UserInfo("user_5","name_5","address_5")
      ))


    /** 2) 流式数据: 用户发的tweet数据*/
    /** 数据示例： * eventTime：事件时间、retweetCount：转推数、language：语言、userID：用户ID、favoriteCount：点赞数、id：事件ID * {"eventTime": "2018-11-05 10:04:00", "retweetCount": 1, "language": "chinese", "userID": "user_1", "favoriteCount": 1, "id": 4909846540155641457} */
    val kafkaDStream=KafkaUtils.createDirectStream[String,String](
      ssc,
      LocationStrategies.PreferConsistent,
      ConsumerStrategies.Subscribe[String,String](List("testTopic3"),kafkaParams)
    ).map(item=>parseJson(item.value())).map(item=>{ 
   
      val userID = item.getString("userID")
      val eventTime = item.getString("eventTime")
      val language= item.getString("language")
      val favoriteCount = item.getInteger("favoriteCount")
      val retweetCount = item.getInteger("retweetCount")
      (userID,(userID,eventTime,language,favoriteCount,retweetCount))
    })


    /** 3) 流与静态数据做Join (Broadcast Join 方式)*/
    val result=kafkaDStream.mapPartitions(part=>{ 
   
      val userInfo = broadcastUserInfo.value
      part.map(item=>{ 
   
        (item._1,(item._2,userInfo.getOrElse(item._1,null)))})
    })

    result.foreachRDD(_.foreach(println))


    ssc.start()
    ssc.awaitTermination()

  }

  /**json解析*/
  def parseJson(log:String):JSONObject={ 
   
    var ret:JSONObject=null
    try{ 
   
      ret=JSON.parseObject(log)
    }catch { 
   
      //异常json数据处理
      case e:JSONException => println(log)
    }
    ret
  }

}

流与半静态数据Join

半静态数据指的是放在Redis等的数据，会被更新。

思路：RDD 每个Partition连接一次Redis,遍历Partition中每条数据，根据k,去Redis中查找v。

package com.bigData.spark

import com.alibaba.fastjson.{ 
   JSON, JSONException, JSONObject}
import org.apache.kafka.common.serialization.StringDeserializer
import org.apache.log4j.{ 
   Level, Logger}
import org.apache.spark.SparkConf
import org.apache.spark.streaming.kafka010.{ 
   ConsumerStrategies, KafkaUtils, LocationStrategies}
import org.apache.spark.streaming.{ 
   Durations, StreamingContext}
import redis.clients.jedis.Jedis

/** * Author: Wang Pei * License: Copyright(c) Pei.Wang * Summary: * * Stream-Static Join * * spark 2.2.2 * */
object StreamStaicJoin3 { 
   
  def main(args: Array[String]): Unit = { 
   

    //设置日志等级
    Logger.getLogger("org").setLevel(Level.WARN)

    //Kafka 参数
    val kafkaParams= Map[String, Object](
      "bootstrap.servers" -> "localhost:9092",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[StringDeserializer],
      "auto.offset.reset" -> "latest",
      "enable.auto.commit" -> (true: java.lang.Boolean),
      "group.id" -> "testTopic3_consumer_v1")

    //spark环境
    val sparkConf = new SparkConf().setAppName(this.getClass.getSimpleName.replace("$","")).setMaster("local[3]")
    val ssc = new StreamingContext(sparkConf,Durations.seconds(10))

    /** 1) 半静态数据: 用户基础信息,在Redis中*/
    /** HMSET user_1 userID "user_1" name "name_1" address "address_1" */
    /** HMSET user_2 userID "user_2" name "name_2" address "address_2" */


    /** 2) 流式数据: 用户发的tweet数据*/
    /** 数据示例： * eventTime：事件时间、retweetCount：转推数、language：语言、userID：用户ID、favoriteCount：点赞数、id：事件ID * {"eventTime": "2018-11-05 10:04:00", "retweetCount": 1, "language": "chinese", "userID": "user_1", "favoriteCount": 1, "id": 4909846540155641457} */

    val kafkaDStream=KafkaUtils.createDirectStream[String,String](
      ssc,
      LocationStrategies.PreferConsistent,
      ConsumerStrategies.Subscribe[String,String](Set("testTopic3"),kafkaParams)
    ).map(item=>parseJson(item.value())).map(item=>{ 
   
      val userID = item.getString("userID")
      val eventTime = item.getString("eventTime")
      val language= item.getString("language")
      val favoriteCount = item.getInteger("favoriteCount")
      val retweetCount = item.getInteger("retweetCount")
      (userID,(userID,eventTime,language,favoriteCount,retweetCount))
    })

    /** 3) 流与半静态数据做Join (RDD Join 方式)*/
    val result=kafkaDStream.mapPartitions(part=>{ 
   
      val redisCli=connToRedis("localhost",6379,3000,10)
      part.map(item=>{ 
   
        (item._1,(item._2,redisCli.hmget(item._1,"userID","name","address")))
      })
    })

    result.foreachRDD(_.foreach(println))


    ssc.start()
    ssc.awaitTermination()

  }

  /**json解析*/
  def parseJson(log:String):JSONObject={ 
   
    var ret:JSONObject=null
    try{ 
   
      ret=JSON.parseObject(log)
    }catch { 
   
      //异常json数据处理
      case e:JSONException => println(log)
    }
    ret
  }

  /**连接到redis*/
  def connToRedis(redisHost:String,redisPort:Int,timeout:Int,dbNum:Int): Jedis ={ 
   
    val redisCli=new Jedis(redisHost,redisPort,timeout)
    redisCli.connect()
    redisCli.select(dbNum)
    redisCli
  }

}

Stream-Stream Join

流与流Join。

思路：DStream Join DStream。

package com.bigData.spark

import com.alibaba.fastjson.{ 
   JSON, JSONException, JSONObject}
import org.apache.kafka.common.serialization.StringDeserializer
import org.apache.log4j.{ 
   Level, Logger}
import org.apache.spark.{ 
   SparkConf, SparkContext}
import org.apache.spark.streaming.kafka010.{ 
   ConsumerStrategies, KafkaUtils, LocationStrategies}
import org.apache.spark.streaming.{ 
   Durations, StreamingContext}

/** * Author: Wang Pei * License: Copyright(c) Pei.Wang * Summary: * * Stream-Stream Join * * spark 2.2.2 * */
object StreamStreamJoin { 
   
  def main(args: Array[String]): Unit = { 
   

    //设置日志等级
    Logger.getLogger("org").setLevel(Level.WARN)

    //Kafka 参数
    val kafkaParams1= Map[String, Object](
      "bootstrap.servers" -> "localhost:9092",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[StringDeserializer],
      "auto.offset.reset" -> "latest",
      "enable.auto.commit" -> (true: java.lang.Boolean),
      "group.id" -> "testTopic3_consumer_v1")

    val kafkaParams2= Map[String, Object](
      "bootstrap.servers" -> "localhost:9092",
      "key.deserializer" -> classOf[StringDeserializer],
      "value.deserializer" -> classOf[StringDeserializer],
      "auto.offset.reset" -> "latest",
      "enable.auto.commit" -> (true: java.lang.Boolean),
      "group.id" -> "testTopic4_consumer_v1")


    //spark环境
    val sparkConf = new SparkConf().setAppName(this.getClass.getSimpleName.replace("$","")).setMaster("local[3]")
    val ssc = new StreamingContext(sparkConf,Durations.seconds(10))

    /** 1) 流式数据: 用户发的tweet数据*/
    /** 数据示例： * eventTime：事件时间、retweetCount：转推数、language：语言、userID：用户ID、favoriteCount：点赞数、id：事件ID * {"eventTime": "2018-11-05 10:04:00", "retweetCount": 1, "language": "chinese", "userID": "user_1", "favoriteCount": 1, "id": 4909846540155641457} */

    val kafkaDStream1=KafkaUtils.createDirectStream[String,String](
      ssc,
      LocationStrategies.PreferConsistent,
      ConsumerStrategies.Subscribe[String,String](List("testTopic3"),kafkaParams1)
    ).map(item=>parseJson(item.value())).map(item=>{ 
   
      val userID = item.getString("userID")
      val eventTime = item.getString("eventTime")
      val language= item.getString("language")
      val favoriteCount = item.getInteger("favoriteCount")
      val retweetCount = item.getInteger("retweetCount")
      (userID,(userID,eventTime,language,favoriteCount,retweetCount))
    })

    /** 2) 流式数据: 用户发的tweet数据*/
    /** 数据示例： * eventTime：事件时间、retweetCount：转推数、language：语言、userID：用户ID、favoriteCount：点赞数、id：事件ID * {"eventTime": "2018-11-05 10:04:00", "retweetCount": 1, "language": "chinese", "userID": "user_1", "favoriteCount": 1, "id": 4909846540155641457} */

    val kafkaDStream2=KafkaUtils.createDirectStream[String,String](
      ssc,
      LocationStrategies.PreferConsistent,
      ConsumerStrategies.Subscribe[String,String](List("testTopic4"),kafkaParams2)
    ).map(item=>parseJson(item.value())).map(item=>{ 
   
      val userID = item.getString("userID")
      val eventTime = item.getString("eventTime")
      val language= item.getString("language")
      val favoriteCount = item.getInteger("favoriteCount")
      val retweetCount = item.getInteger("retweetCount")
      (userID,(userID,eventTime,language,favoriteCount,retweetCount))
    })

    /** 3) Stream-Stream Join*/
    val joinedDStream = kafkaDStream1.leftOuterJoin(kafkaDStream2)

    joinedDStream.foreachRDD(_.foreach(println))

    ssc.start()
    ssc.awaitTermination()

  }

  /**json解析*/
  def parseJson(log:String):JSONObject={ 
   
    var ret:JSONObject=null
    try{ 
   
      ret=JSON.parseObject(log)
    }catch { 
   
      //异常json数据处理
      case e:JSONException => println(log)
    }
    ret
  }

}