场景
餐厅老板想要统计每个用户来他的店里总共消费了多少金额,我们可以使用updateStateByKey来实现
从kafka接收用户消费json数据,统计每分钟用户的消费情况,并且统计所有时间所有用户的消费情况(使用updateStateByKey来实现)
数据格式
{"user":"zhangsan","payment":8}
{"user":"wangwu","payment":7}
....
往kafka写入消息(kafka producer)
package producer
import java.util.Properties
import kafka.javaapi.producer.Producer
import kafka.producer.{KeyedMessage, ProducerConfig}
import org.codehaus.jettison.json.JSONObject
import scala.util.Random
object KafkaProducer extends App{
//所有用户
private val users = Array(
"zhangsan", "lisi",
"wangwu", "zhaoliu")
private val random = new Random()
//消费的金额(0-9)
def payMount() : Double = {
random.nextInt(10)
}
//随机获得用户名称
def getUserName() : String = {
users(random.nextInt(users.length))
}
//kafka参数
val topic = "user_payment"
val brokers = "192.168.6.55:9092,192.168.6.56:9092"
val props = new Properties()
props.put("metadata.broker.list", brokers)
props.put("serializer.class", "kafka.serializer.StringEncoder")
val kafkaConfig = new ProducerConfig(props)
val producer = new Producer[String, String](kafkaConfig)
while(true) {
// 创建json串
val event = new JSONObject()
event
.put("user", getUserName())
.put("payment", payMount)
// 往kafka发送数据
producer.send(new KeyedMessage[String, String](topic, event.toString))
println("Message sent: " + event)
//每隔200ms发送一条数据
Thread.sleep(200)
}
}
使用spark Streaming处理数据
import org.apache.spark.streaming.kafka.KafkaUtils
import org.apache.spark.streaming.{StreamingContext, Seconds}
import org.apache.spark.{SparkContext, SparkConf}
import net.liftweb.json._
object UpdateStateByKeyTest {
def main (args: Array[String]) {
def functionToCreateContext(): StreamingContext = {
//创建streamingContext
val conf = new SparkConf().setAppName("test").setMaster("local[*]")
val ssc = new StreamingContext(conf, Seconds(60))
//将数据进行保存(这里作为演示,生产中保存在hdfs)
ssc.checkpoint("checkPoint")
val zkQuorum = "192.168.6.55:2181,192.168.6.56:2181,192.168.6.57:2181"
val consumerGroupName = "user_payment"
val kafkaTopic = "user_payment"
val kafkaThreadNum = 1
val topicMap = kafkaTopic.split(",").map((_, kafkaThreadNum.toInt)).toMap
//从kafka读入数据并且将json串进行解析
val user_payment = KafkaUtils.createStream(ssc, zkQuorum, consumerGroupName, topicMap).map(x=>{
parse(x._2)
})
//对一分钟的数据进行计算
val paymentSum = user_payment.map(jsonLine =>{
implicit val formats = DefaultFormats
val user = (jsonLine \ "user").extract[String]
val payment = (jsonLine \ "payment").extract[String]
(user,payment.toDouble)
}).reduceByKey(_+_)
//输出每分钟的计算结果
paymentSum.print()
//将以前的数据和最新一分钟的数据进行求和
val addFunction = (currValues : Seq[Double],preVauleState : Option[Double]) => {
val currentSum = currValues.sum
val previousSum = preVauleState.getOrElse(0.0)
Some(currentSum + previousSum)
}
val totalPayment = paymentSum.updateStateByKey[Double](addFunction)
//输出总计的结果
totalPayment.print()
ssc
}
//如果"checkPoint"中存在以前的记录,则重启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()
}
}
运行结果节选
//-----------第n分钟的结果------------------
//1分钟结果
-------------------
(zhangsan,23.0)
(lisi,37.0)
(wangwu,31.0)
(zhaoliu,34.0)
-------------------
//总和结果
(zhangsan,101.0)
(lisi,83.0)
(wangwu,80.0)
(zhaoliu,130.0)
//-----------第n+1分钟的结果------------------
//1分钟结果
-------------------
(zhangsan,43.0)
(lisi,16.0)
(wangwu,21.0)
(zhaoliu,54.0)
-------------------
//总和结果
-------------------
(zhangsan,144.0)
(lisi,99.0)
(wangwu,101.0)
(zhaoliu,184.0)
-------------------
后记
下一片文章为统计不同时间段用户平均消费金额,消费次数,消费总额等指标。
点击这里