SUM OVER PARTITION BY | 易学教程

问题

What am I missing?

This query is returning duplicate data over and over again. The count is correct for a complete total, but I am expecting one row, and yet I am getting the value repeated about 40 times. Any ideas?

SELECT BrandId
      ,SUM(ICount) OVER (PARTITION BY BrandId ) 
  FROM Table 
WHERE DateId  = 20130618

I get this?

BrandId ICount
2       421762
2       421762
2       421762
2       421762
2       421762
2       421762
2       421762
1       133346
1       133346
1       133346
1       133346
1       133346
1       133346
1       133346

What am I missing?

I cant remove the partition by as the entire query is like this:

SELECT BrandId
       ,SUM(ICount) OVER (PARTITION BY BrandId) 
       ,TotalICount= SUM(ICount) OVER ()    
        ,SUM(ICount) OVER () / SUM(ICount) OVER (PARTITION BY BrandId)  as Percentage
FROM Table 
WHERE DateId  = 20130618

Which returns this:

BrandId (No column name)    TotalICount Percentage
2       421762              32239892    76
2       421762              32239892    76
2       421762              32239892    76
2       421762              32239892    76
2       421762              32239892    76
2       421762              32239892    76

I would expect output something like this without having to use a distinct:

BrandId (No column name)    TotalICount Percentage
2       421762              32239892    76
9       1238442             32239892    26
10      1467473             32239892    21

回答1:

You could have used DISTINCT or just remove the PARTITION BY portions and use GROUP BY:

SELECT BrandId
       ,SUM(ICount)
       ,TotalICount = SUM(ICount) OVER ()    
       ,Percentage = SUM(ICount) OVER ()*1.0 / SUM(ICount) 
FROM Table 
WHERE DateId  = 20130618
GROUP BY BrandID

Not sure why you are dividing the total by the count per BrandID, if that's a mistake and you want percent of total then reverse those bits above to:

SELECT BrandId
           ,SUM(ICount)
           ,TotalICount = SUM(ICount) OVER ()    
           ,Percentage = SUM(ICount)*1.0 / SUM(ICount) OVER () 
    FROM Table 
    WHERE DateId  = 20130618
    GROUP BY BrandID

回答2:

In my opinion, I think it's important to explain the why behind the need for a GROUP BY in your SQL when summing with OVER() clause and why you are getting repeated lines of data when you are expecting one row per BrandID.

Take this example: You need to aggregate the total sale price of each order line, per specific order category, between two dates, but you also need to retain individual order data in your final results. A SUM() on the SalesPrice column would not allow you to get the correct totals because it would require a GROUP BY, therefore squashing the details because you wouldn't be able to keep the individual order lines in the select statement.

Many times we see a #temp table, @table variable, or CTE filled with the sum of our data and grouped up so we can join to it again later to get a column of the sums we need. This can add processing time and extra lines of code. Instead, use OVER(PARTITION BY ()) like this:

SELECT
  OrderLine, 
  OrderDateTime, 
  SalePrice, 
  OrderCategory,
  SUM(SalePrice) OVER(PARTITION BY OrderCategory) AS SaleTotalPerCategory
FROM tblSales 
WHERE OrderDateTime BETWEEN @StartDate AND @EndDate

Notice we are not grouping and we have individual order lines column selected. The PARTITION BY in the last column will return us a sales price total for each row of data in each category. What the last column essentially says is, we want the sum of the sale price (SUM(SalePrice)) over a partition of my results and by a specified category (OVER(PARTITION BY CategoryHere)).

If we remove the other columns from our select statement, and leave our final SUM() column, like this:

SELECT
  SUM(SalePrice) OVER(PARTITION BY OrderCategory) AS SaleTotalPerCategory
FROM tblSales 
WHERE OrderDateTime BETWEEN @StartDate AND @EndDate

The results will still repeat this sum for each row in our original result set. The reason is this method does not require a GROUP BY. If you don't need to retain individual line data, then simply SUM() without the use of OVER() and group up your data appropriately. Again, if you need an additional column with specific totals, you can use the OVER(PARTITION BY ()) method described above without additional selects to join back to.

The above is purely for explaining WHY he is getting repeated lines of the same number and to help understand what this clause provides. This method can be used in many ways and I highly encourage further reading from the documentation here:

Over Clause

回答3:

I think the query you want is this:

SELECT BrandId, SUM(ICount),
       SUM(sum(ICount)) over () as TotalCount,
       100.0 * SUM(ICount) / SUM(sum(Icount)) over () as Percentage
FROM Table 
WHERE DateId  = 20130618
group by BrandId;

This does the group by for brand. And it calculates the "Percentage". This version should produce a number between 0 and 100.

回答4:

remove partition by and add group by clause,

SELECT BrandId
      ,SUM(ICount) totalSum
  FROM Table 
WHERE DateId  = 20130618
GROUP BY BrandId

来源：https://stackoverflow.com/questions/17867807/sum-over-partition-by

标签

sql

sql-server

tsql