2
有两种方法可以在两个并发的Spark作业中使用相同的RDD吗?如何在并发Spark作业中共享RDD
例如,在下面的应用程序中,我试图将b
写入磁盘(作业1),同时计算f
(作业2)。但是,Spark似乎一次只能执行一项工作。
val conf = new SparkConf()
val sc = new SparkContext(conf)
val a = sc.parallelize(0 until 1000)
val b = a.mapPartitions(it => { Thread.sleep(5000); it })
// Compute b
b.persist().foreachPartition(_ => {})
val c = b.mapPartitions(it => { Thread.sleep(5000); it })
val d = c.mapPartitions(it => { Thread.sleep(5000); it })
val e = d.mapPartitions(it => { Thread.sleep(5000); it })
val f = e.mapPartitions(it => { Thread.sleep(5000); it })
// Concurrent actions on b and f (f uses b)
val actionFuts = List(
// Job 1
Future {
Thread.sleep(1000)
b.saveAsTextFile("output.ignore/test/b.txt")
},
// Job 2
Future {
f.saveAsTextFile("output.ignore/test/f.txt")
}
)
Await.result(Future.sequence(actionFuts).map(_ =>()), Duration.Inf)