by Mark Wills
PIVOT is a great facility and solves many an EAV (Entity - Attribute - Value) type transformation where we need the information held as data within a column to become columns in their own right. Now, in some cases that is relatively easy to do, other times it can be a challenge (and that is why we are here).
Let's have a quick look at the PIVOT function...
SELECT
<display_column_list>
FROM
(SELECT <source_columns> as Column_Source
,<column_to_be_aggregated> as Column_Value
,<column_with_new_column_names> as Column_List
FROM <datasource> ) as DataSource
PIVOT
(<aggregate_function>(Column_Value) FOR Column_List IN
([<new_column_1_heading>],[<new_column_2_heading>],...,[<new_column_N_heading>]) ) PivotTable
ORDER BY <column_number_or_name>;
That looks pretty straight forward, except for one or two small details:
1) First up, we need to know the <display_column_list>
easy enough, just do a Select * instead and problem solved (except that it does control display sequence)
2) Secondly, we need to know and hard code the new column headings as in :
([<new_column_1_heading>],[<new_column_2_heading>],...,[<new_column_N_heading>])
And that last point is often the big challenge. Not so bad if we are doing something static like "months in a year", just list out those months in sequence, and make sure you can cast the column which contains those new headings accordingly e.g. datename (month,<column_with_new_column_names>)
But, what about a moving target - like "last 3 months" ? Or, an EAV table with unknown attribute names ? Normally, that means we need to rewrite our query every month, or after every change of data.
There is a way, and that involves some Dynamic SQL, more importantly, we can make it a procedure which can handle any "simple" dynamic pivot table.
So, lets get started... but first we need a fairly simple example, so we will create some data accordingly. Feeling generous, we will do two. One is a classic "rolling periods" and the other a typical EAV
CREATE TABLE tst_CustSales (
TCS_ID INT Identity Primary Key Clustered,
TCS_Customer varchar(60),
TCS_Date DATETIME,
TCS_Quantity INT,
TCS_Value MONEY )
GO
CREATE TABLE tst_EAV_Data (
TED_ID INT Identity Primary Key Clustered,
TED_Entity varchar(60),
TED_Attribute varchar(60),
TED_Value varchar(60) )
GO
-- now let's populate our tst_* tables
INSERT tst_CustSales (TCS_Customer, TCS_Date, TCS_Quantity, TCS_Value)
SELECT * FROM (
SELECT 'Customer 1' as Customer,'20090101' as Date, 11 as Qty, 1001.00 as Val union all
SELECT 'Customer 1','20090201',12, 1002.00 union all
SELECT 'Customer 1','20090301',13, 1003.00 union all
SELECT 'Customer 1','20090401',14, 1004.00 union all
SELECT 'Customer 2','20090101',21, 2001.00 union all
SELECT 'Customer 2','20090201',22, 2002.00 union all
SELECT 'Customer 2','20090301',23, 2003.00 union all
SELECT 'Customer 2','20090401',24, 2004.00 union all
SELECT 'Customer 3','20090101',31, 3001.00 union all
SELECT 'Customer 4','20090201',42, 4002.00 union all
SELECT 'Customer 5','20090301',53, 5003.00 ) as src
GO
-- notice I do not mention the Identity Column - SQL will manage that for me
-- notice the yyyymmdd "style 112" format - implicitly converts to datetime
-- now our EAV table, again imagine some diverse attributes
INSERT tst_EAV_Data (TED_Entity, TED_Attribute, TED_Value)
SELECT * FROM (
SELECT 'Customer 1' as Customer,'Phone' as Attr,'+61299991234' as Data_Val union all
SELECT 'Customer 1','Address','24 Somewhere Street' union all
SELECT 'Customer 1','Building','The ReallyTall One' union all
SELECT 'Customer 1','Contact','Marcus Aurelius' union all
SELECT 'Customer 2','Phone','+61288881234' union all
SELECT 'Customer 2','Contact','Ritesh Shah' union all
SELECT 'Customer 3','Address','1600 Pennsylvania Avenue' union all
SELECT 'Customer 3','Building','The WhiteHouse' union all
SELECT 'Customer 4','Phone','+61277771234' union all
SELECT 'Customer 4','Address','1 Nile Way' union all
SELECT 'Customer 4','Building','The Pyramids' union all
SELECT 'Customer 4','Contact','Cleo Patra' union all
SELECT 'Customer 5','Phone','+61277771222' union all
SELECT 'Customer 5','Friend','Cleo Patra' ) as src
GO
1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:
27:
28:
29:
30:
31:
32:
33:
34:
35:
36:
37:
38:
39:
40:
41:
42:
43:
44:
45:
46:
47:
48:
49:
50:
51:
52:
53:
Select allOpen in new window
Now we can get down and dirty with the Pivot.
First we will construct a properly formed one so you can "see" the pivot in action.
SELECT TCS_Customer, [01 Feb 2009],[01 Mar 2009],[01 Apr 2009]
FROM
(select TCS_Customer, TCS_Date, TCS_Value from tst_CustSales ) sourcedata
PIVOT
(sum(TCS_Value) for TCS_Date in ([01 Feb 2009],[01 Mar 2009],[01 Apr 2009])) pivottable
GO
1:
2:
3:
4:
5:
6:
Select allOpen in new window
You can see from the above that the column_list and headings are all hard coded...
Also note how SQL is dynamically converting the datetime to those column headings that is because dd MMM yyyy can be implicitly converted to datetime when used in a date context. But "Style 106" is language dependant, so you do need to take care.
In this case, amazingly, can handle the "hard coded" column names (note we have no time component in our test data, we show you below how to remove time).
Now let us have a look at some Dynamic SQL for the EAV table, again in "long hand". The dynamic bit is getting those column names so we do not have to hard code them...
DECLARE @Columns varchar(8000)
DECLARE @SQL varchar(8000)
SET @Columns = substring((select ',['+TED_Attribute+']' from tst_EAV_Data group by TED_Attribute for xml path('')),2,8000)
SET @SQL = 'SELECT * FROM
(Select TED_Entity as Cust,TED_Attribute,TED_Value from tst_EAV_Data) sourcedata
PIVOT
(max(TED_Value) for TED_Attribute in ('+@Columns+')) pivottable'
EXEC(@sql)
GO
1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
11:
Select allOpen in new window
Let's have a look at the above, all we really did was to generate the column list.
You can try it again replacing the EXEC(@SQL) with Print @SQL
You will see pretty much the same command structure as the earlier pivot.
Now to create a Procedure so we can simply keep using a stored procedure rather than having to write code all the time.
So lets get into it...
CREATE PROCEDURE uDynamicPivot(
@sourcedata varchar(8000),
@Pivot_On_Source_Column varchar(2000),
@Pivot_Value_Aggregate varchar(10),
@Pivot_Value_Column varchar(2000),
@Pivot_Column_List varchar(2000),
@Pivot_Column_Style_Code varchar(4)) -- used in convert for style code
AS
BEGIN
-- we really should put in some error checking, e.g. if anything is NULL it will crash.
declare @columns varchar(max)
declare @sql nvarchar(max)
set @sql = N'set @columns = substring((select '', [''+convert(varchar,'+@Pivot_Column_List+@Pivot_Column_Style_Code+')+'']'' from '+@sourcedata+' group by '+@Pivot_Column_List+' for xml path('''')),2,8000)'
execute sp_executesql @sql,
N'@columns varchar(max) output',
@columns=@columns output
set @sql = N'SELECT * FROM
(SELECT '+@Pivot_On_Source_Column+','+@Pivot_Column_List+','+@Pivot_Value_Column+' from '+@sourcedata+') src
PIVOT
('+@Pivot_Value_Aggregate+'('+@Pivot_Value_Column+') FOR '+@Pivot_Column_List+' IN ('+@columns+') ) pvt
ORDER BY 1'
execute sp_executesql @sql
END
GO
1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:
27:
28:
Select allOpen in new window
Now, let's use that procedure by plugging the parameters needed for a PIVOT function
uDynamicPivot 'tst_CustSales','TCS_customer','sum','TCS_Value','TCS_Date',',106'
and the EAV
uDynamicPivot 'tst_EAV_Data','TED_Entity as Cust','max','TED_Value','TED_Attribute',''
We can even include some "where" clauses for simple requirements
uDynamicPivot 'tst_CustSales where TCS_Date >= convert(varchar(6),dateadd(month,-3,getdate()),112)+''01''','TCS_customer','sum','TCS_Value','TCS_Date',',106'
But that is getting pretty ugly, and that is where the VIEW comes into play...
VIEWS allow data to be presented in a similar way in which we use a table.
A view is a good way to present data that does need some kind of transformation. It also allows a certain detachment from the underlying table.
Views are really a pointer or script to the actual data and does not contain data itself more so the "rules" on how to get/show the data.
Once created, it is part of the database and can be re-used as often as you like.
CREATE VIEW vw_last_3_months AS
SELECT TCS_Customer as Customer
, TCS_Value
, DATEADD(day, DATEDIFF(day, 0, TCS_Date),0) as Date
, DATEADD(month, DATEDIFF(month, 0, TCS_Date),0) as Start_Of_Month
, DATEADD(month, DATEDIFF(month, -1, TCS_Date), -1) as End_Of_Month
FROM tst_CustSales
WHERE TCS_Date >= convert(varchar(6),dateadd(month,-3,getdate()),112)+'01'
GO
1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
Select allOpen in new window
Note how we are using date functions to transform our date data to remove time and generate a start and end of month.
Now we can do our "simple" function call using our View.
uDynamicPivot 'vw_last_3_months','customer','sum','TCS_Value','End_Of_Month',',106'
And that as they say is that.
Please take care when running on your machine make sure you check table names, double check your code, go step at a time. And hope you have some fun with it.
by: DanielWilson on 2009-06-09 at 10:06:42ID: 1493
Very cool, Mark.
The downside of having it use dynamic SQL is worth it to have a generic solution like this. This is definitely one I will keep in mind.