下面的操作会影响到Spark输出RDD分区(partitioner)的:
cogroup, groupWith, join, leftOuterJoin, rightOuterJoin, groupByKey, reduceByKey, combineByKey, partitionBy, sort, mapValues (如果父RDD存在partitioner), flatMapValues(如果父RDD存在partitioner), 和 filter (如果父RDD存在partitioner)。其他的transform操作不会影响到输出RDD的partitioner,一般来说是None,也就是没有partitioner。下面举个例子进行说明:
scala> val pairs = sc.parallelize(List((1, 1), (2, 2), (3, 3))) pairs: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[4] at parallelize at <console>:12 scala> val a = sc.parallelize(List(2,51,2,7,3)) a: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[5] at parallelize at <console>:12 scala> val a = sc.parallelize(List(2,51,2)) a: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[6] at parallelize at <console>:12 scala> val b = sc.parallelize(List(3,1,4)) b: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[7] at parallelize at <console>:12 scala> val c = a.zip(b) c: org.apache.spark.rdd.RDD[(Int, Int)] = ZippedPartitionsRDD2[8] at zip at <console>:16 scala> val result = pairs.join(c) result: org.apache.spark.rdd.RDD[(Int, (Int, Int))] = FlatMappedValuesRDD[11] at join at <console>:20 scala> result.partitioner res6: Option[org.apache.spark.Partitioner] = Some(org.apache.spark.HashPartitioner@2)
大家可以看到输出来的RDD result分区变成了HashPartitioner,因为join中的两个分区都没有设置分区,所以默认用到了HashPartitioner,可以看join的实现:
def join[W](other: RDD[(K, W)]): RDD[(K, (V, W))] = { join(other, defaultPartitioner(self, other)) } def defaultPartitioner(rdd: RDD[_], others: RDD[_]*): Partitioner = { val bySize = (Seq(rdd) ++ others).sortBy(_.partitions.size).reverse for (r <- bySize if r.partitioner.isDefined) { return r.partitioner.get } if (rdd.context.conf.contains("spark.default.parallelism")) { new HashPartitioner(rdd.context.defaultParallelism) } else { new HashPartitioner(bySize.head.partitions.size) } }
defaultPartitioner
函数就确定了结果RDD的分区。从上面的实现可以看到,
1、join的两个RDD如果都没有partitioner,那么join结果RDD将使用HashPartitioner;
2、如果两个RDD中其中有一个有partitioner,那么join结果RDD将使用那个父RDD的partitioner;
3、如果两个RDD都有partitioner,那么join结果RDD就使用调用join的那个RDD的partitioner。
原创文章版权归过往记忆大数据(过往记忆)所有,未经许可不得转载。
本文链接: 【影响到Spark输出RDD分区的操作函数】(https://www.iteblog.com/archives/1242.html)