Spark, Hive SQL - implementing window function?

Question

I'm trying to implement the following solution: window function

I have the following 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|                                 
+------------+----------------------+-------------------+

I'm trying to run a window function as:

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();

What I want is to get the last two entries (last as for the latest date, but since it's ordered by date descending, the first two rows) for each user:

+------------+----------------------+-------------------+                                 
|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|                                     
+------------+----------------------+-------------------+

but I get this:

+------------+----------------------+-------------------+----+
|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|                            
+------------+----------------------+-------------------+----+

What am I missing?

score 0 · Accepted Answer · answered Jul 20 '16 at 12:08

0

All values are equal -> ranks are equal. Try row_number:

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

answered Jul 20 '16 at 12:08

Thanks! This works perfectly! So it also seems the original one would work for real-life data, too, for the timestamps would be different. :) – lte__ Jul 20 '16 at 15:24

Spark, Hive SQL - implementing window function?

1 Answers1