通过sc.broadcast广播出去
val broadcastVar = sc.broadcast(Array(1, 2, 3))
通过broadcastVar.value取到值
broadcastVar.value
这是一个join案例,通过广播变量以降低通信成本
广播变量是把小表的数据通过sc广播出去
def main(args: Array[String]): Unit = {
val spark = SparkSession.builder().master("local").appName("JoinApp").getOrCreate()
// 将小表数据获得一个map,这边通过collectAsMap()将rdd转成map
val peopleInfo: collection.Map[String, String] = spark.sparkContext
.parallelize(Array(("100", "pk"), ("101", "jepson"))).collectAsMap()
//通过spark.sparkContext.broadcast(peopleInfo)将其广播出去
val peopleBroadcast: Broadcast[collection.Map[String, String]] = spark.sparkContext.broadcast(peopleInfo)
//获得大表数据,将其转化成键值对型的rdd数据
val peopleDetail: RDD[(String, (String, String, String))] = spark.sparkContext
.parallelize(Array(("100", "ustc", "beijing"), ("103", "xxx", "shanghai"),("101", "xxx", "guangzhou")))
.map(x => (x._1, x))
//使用mapPartitions遍历处理批数据
// mapPartitions做的事情: 遍历大表的每一行数据 和 广播变量的数据对比 有就取出来,没有就拉倒
val value: RDD[(String, String, String)] = peopleDetail.mapPartitions(x => {
//通过peopleBroadcast.value获得map类型的小表数据
val broadcastPeople: collection.Map[String, String] = peopleBroadcast.value
//遍历小表数据,判断大表key是否在小表的key中存在
for((key,value) <- x if broadcastPeople.contains(key))
//for循环的返回值一个可迭代元组,将数据返回
yield (key, broadcastPeople.get(key).getOrElse(""), value._2)
})
value.foreach(println)
spark.stop()
}
查看控制台