Spark 重分区函数：coalesce和repartition区别与实现，可以优化Spark程序性能

Posted 2023-02-02 javartisan

源码包路径： org.apache.spark.rdd.RDD

coalesce函数：

方法注释：

Return a new RDDthat is reduced into numPartitions partitions. This results in a narrowdependency, e.g. if you go from 1000 partitions to 100 partitions, there willnot be a shuffle, instead each of the 100 new partitions will claim 10 of thecurrent partitions. However, if you're doing a drastic(激烈的,猛烈的)coalesce, e.g. to numPartitions = 1, this may result in your computation takingplace on fewer nodes than you like (e.g. one node in the case of numPartitions= 1). To avoid this, you can pass shuffle = true. This will add a shuffle step,but means the current upstream partitions will be executed in parallel (perwhatever the current partitioning is). Note: With shuffle = true, you canactually coalesce to a larger number of partitions. This is useful if you have asmall number of partitions, say 100, potentially with a few partitions beingabnormally large. Calling coalesce(1000, shuffle = true) will result in 1000partitions with the data distributed using a hash partitioner.

 

译文：

返回一个经过简化到numPartitions个分区的新RDD。这会导致一个窄依赖，例如：你将1000个分区转换成100个分区，这个过程不会发生shuffle，相反如果10个分区转换成100个分区将会发生shuffle。然而如果你想大幅度合并分区，例如合并成一个分区，这会导致你的计算在少数几个集群节点上计算（言外之意：并行度不够）。为了避免这种情况，你可以将第二个shuffle参数传递一个true，这样会在重新分区过程中多一步shuffle，这意味着上游的分区可以并行运行。

 

注意：第二个参数shuffle=true，将会产生多于之前的分区数目，例如你有一个个数较少的分区，假如是100，调用coalesce(1000, shuffle = true)将会使用一个  HashPartitioner产生1000个分区分布在集群节点上。这个（对于提高并行度）是非常有用的。

repartition函数：

Return a new RDDthat has exactly numPartitions partitions. Can increase or decrease the levelof parallelism in this RDD. Internally, this uses a shuffle to redistributedata. If you are decreasing the number of partitions in this RDD, considerusing coalesce, which can avoid performing a shuffle.

译文：

返回一个恰好有numPartitions个分区的RDD，可以增加或者减少此RDD的并行度。内部，这将使用shuffle重新分布数据，如果你减少分区数，考虑使用coalesce，这样可以避免执行shuffle