Sum Consecutive Months Based on Groups with Criteria - sql-server

Question

I am having trouble narrowing down sales in top regions that occurred in consecutive months. I know I need to use some form of window function with Row_Number or Dense_Rank, but I am having trouble getting the final output
Here is my source data:
+--------+-----------+------------+
| Fruit | SaleDate | Top_Region |
+--------+-----------+------------+
| Apple | 1/1/2017 | 1 |
| Apple | 2/1/2017 | 1 |
| Apple | 3/1/2017 | 1 |
| Apple | 4/1/2017 | 0 |
| Apple | 5/1/2017 | 0 |
| Apple | 6/1/2017 | 0 |
| Apple | 7/1/2017 | 1 |
| Apple | 8/1/2017 | 1 |
| Apple | 9/1/2017 | 1 |
| Apple | 10/1/2017 | 1 |
| Apple | 11/1/2017 | 0 |
| Apple | 12/1/2017 | 0 |
| Banana | 1/1/2017 | 0 |
| Banana | 2/1/2017 | 0 |
| Banana | 3/1/2017 | 1 |
| Banana | 4/1/2017 | 1 |
| Banana | 5/1/2017 | 1 |
| Banana | 6/1/2017 | 1 |
| Banana | 7/1/2017 | 1 |
| Banana | 8/1/2017 | 1 |
| Banana | 9/1/2017 | 0 |
| Banana | 10/1/2017 | 1 |
| Banana | 11/1/2017 | 1 |
| Banana | 12/1/2017 | 0 |
+--------+-----------+------------+
This is the expected output:
+--------+-----------+-----------+-------+
| Fruit | Start | End | Total |
+--------+-----------+-----------+-------+
| Apple | 1/1/2017 | 3/1/2017 | 3 |
| Apple | 7/1/2017 | 10/1/2017 | 4 |
| Banana | 3/1/2017 | 8/1/2017 | 6 |
| Banana | 10/1/2017 | 11/1/2017 | 2 |
+--------+-----------+-----------+-------+
The goal is to have instances of top region sales in succession with missing in one month.
So far I have tried a few different combinations, with this being the closest.
SELECT fruit,
MIN(saledate) AS spanStart ,
MAX(saledate) AS spanEnd,
COUNT(*) AS spanLength
FROM ( SELECT s.* ,
( ROW_NUMBER() OVER ( ORDER BY month )
- ROW_NUMBER() OVER ( PARTITION BY fruit, topregion ORDER BY month ) ) AS fruits
FROM #salesdata s
) s
GROUP BY fruit,fruits ,
topregion
HAVING topregion = 1
ORDER BY COUNT(*) DESC;
Any help would be greatly appreciated

Answer 1

This is a typical gaps-and-island problem. One strategy is to identify the groups of adjacent rows groups by computing the difference between two row_number()s. We can then filter on groups having top_region = 1 and use aggregation to get the start date, end date and number of records per group.
Your query is really close, but the first row_number() is missing a partition by fruit in its over() clause. And I find that aliasing that column fruits where another column is called fruit is error prone.
select
fruit,
min(sale_date) start_date,
max(sale_date) end_date,
count(*) total
from (
select
t.*,
row_number() over(partition by fruit order by sale_date) rn1,
row_number() over(partition by fruit, top_region order by sale_date) rn2
from mytable t
) t
where top_region = 1
group by fruit, rn1 - rn2
order by fruit, start_date
You can run the inner query separately to see the result it produces.
Demo on DB Fiddle:
fruit | start_date | end_date | total
:----- | :--------- | :--------- | ----:
Apple | 2017-01-01 | 2017-01-03 | 3
Apple | 2017-01-07 | 2017-01-10 | 4
Banana | 2017-01-03 | 2017-01-08 | 6
Banana | 2017-01-10 | 2017-01-11 | 2