drop_duplicate是否保证在Spark中对dataframe进行排序后保留第一行并删除其余行？

Question

frame = frame.orderBy(["b","c"],ascending=False)
frame = frame.drop_duplicate('a')

基于Spark Scala代码，我可以看到Orderby在内部调用sort方法，该方法执行全局排序。

/**
   * Returns a new Dataset sorted by the given expressions. For example:
   * {{{
   *   ds.sort($"col1", $"col2".desc)
   * }}}
   *
   * @group typedrel
   * @since 2.0.0
   */
  @scala.annotation.varargs
  def sort(sortExprs: Column*): Dataset[T] = {
    sortInternal(global = true, sortExprs)
  }

https://github.com/apache/spark/blob/branch-2.4/sql/core/src/main/scala/org/apache/spark/sql/dataset.scala

而且drop_duplicates(cols)方法也按照bellow spark代码转换为聚合（first（cols））。

object ReplaceDeduplicateWithAggregate extends Rule[LogicalPlan] {
  def apply(plan: LogicalPlan): LogicalPlan = plan transformUpWithNewOutput {
    case d @ Deduplicate(keys, child) if !child.isStreaming =>
      val keyExprIds = keys.map(_.exprId)
      val aggCols = child.output.map { attr =>
        if (keyExprIds.contains(attr.exprId)) {
          attr
        } else {
          Alias(new First(attr).toAggregateExpression(), attr.name)()
        }
      }
      // SPARK-22951: Physical aggregate operators distinguishes global aggregation and grouping
      // aggregations by checking the number of grouping keys. The key difference here is that a
      // global aggregation always returns at least one row even if there are no input rows. Here
      // we append a literal when the grouping key list is empty so that the result aggregate
      // operator is properly treated as a grouping aggregation.
      val nonemptyKeys = if (keys.isEmpty) Literal(1) :: Nil else keys
      val newAgg = Aggregate(nonemptyKeys, aggCols, child)
      val attrMapping = d.output.zip(newAgg.output)
      newAgg -> attrMapping
  }
}

有什么想法为什么？

Answer 1

没有。

按b&c排序，然后按a删除，如果且仅当只有1个分区要处理，就可以按照您的意愿工作。大数据通常不是这样。

因此，您可以在其他地方谷歌:dropduplicates保留排序操作的第一次出现--只有在有1个分区的情况下，否则就是运气。