Spark，Hive SQL - 实现窗口功能？

Question

我正在尝试实施以下解决方案：window function

我有以下df：

+------------+----------------------+-------------------+                                 
|increment_id|base_subtotal_incl_tax|          eventdate|                                 
+------------+----------------------+-------------------+                                 
|        1086|            14470.0000|2016-06-14 09:54:12|                                 
|        1086|            14470.0000|2016-06-14 09:54:12|                                 
|        1086|            14470.0000|2015-07-14 09:54:12|                                 
|        1086|            14470.0000|2015-07-14 09:54:12|                                 
|        1086|            14470.0000|2015-07-14 09:54:12|                                 
|        1086|            14470.0000|2015-07-14 09:54:12|                                 
|        1086|             1570.0000|2015-07-14 09:54:12|                                 
|        5555|            14470.0000|2014-07-14 09:54:12|                                 
|        5555|            14470.0000|2014-07-14 09:54:12|                                 
|        5555|            14470.0000|2014-07-14 09:54:12|                                 
|        5555|            14470.0000|2014-07-14 09:54:12|                                 
+------------+----------------------+-------------------+ 

我正在尝试将窗口函数运行为：

WindowSpec window = Window.partitionBy(df.col("id")).orderBy(df.col("eventdate").desc());
df.select(df.col("*"),rank().over(window).alias("rank")) //error for this line
         .filter("rank <= 2")
         .show();

我想要的是为每个用户获取最后两个条目（最后一个是最新日期，但由于它是按日期降序排列的，前两行）：

+------------+----------------------+-------------------+                                 
|increment_id|base_subtotal_incl_tax|          eventdate|                                 
+------------+----------------------+-------------------+                                 
|        1086|            14470.0000|2016-06-14 09:54:12|                                 
|        1086|            14470.0000|2016-06-14 09:54:12|   
|        5555|            14470.0000|2014-07-14 09:54:12|                                 
|        5555|            14470.0000|2014-07-14 09:54:12|                                     
+------------+----------------------+-------------------+ 

但我明白了：

+------------+----------------------+-------------------+----+
|increment_id|base_subtotal_incl_tax|          eventdate|rank|                            
+------------+----------------------+-------------------+----+                            
|        5555|            14470.0000|2014-07-14 09:54:12|   1|                            
|        5555|            14470.0000|2014-07-14 09:54:12|   1|                            
|        5555|            14470.0000|2014-07-14 09:54:12|   1|                            
|        5555|            14470.0000|2014-07-14 09:54:12|   1|                            
|        1086|            14470.0000|2016-06-14 09:54:12|   1|                            
|        1086|            14470.0000|2016-06-14 09:54:12|   1|                            
+------------+----------------------+-------------------+----+

我错过了什么？

Answer 1

所有值都相等 -> 等级相等。试试row_number：

df.select(df.col("*"),row_number().over(window).alias("rank"))
     .filter("rank <= 2")
     .show();