One of the good things about SQL Server (as with most other Microsoft products) is that there are often multiple ways to achieve a single result. One such example is the concatenation of cell values from multiple rows into a single, delimited string.
Straight off the bat, people will suggest things like:
While these are all perfectly valid solutions, as always with SQL Server it’s important to investigate the optimum solution for your particular need. After attempting all of the above and encountering some severe performance degradation as number of rows increased, one solution provided the ideal functionality with a huge performance boost (around 1800 times faster): the .NET common language runtime (CLR).
Simply put, CLR integration enables you to write .NET code which can then be called from within User Defined Functions, Stored Procedures and Triggers. You get access to the .NET Framework’s library of built in functions as well as the benefits of the object oriented model. Managed code is also much faster at performing complex calculations and logical operations than T-SQL.
Using Andy Novick’s excellent concatenate aggregate function, concatenating a series of values becomes as simple as a single function call in a SQL script. My requirement was to be able to select the last n number of data points per group, and concatenate these into a single, comma delimited string. Using this CLR function, it becomes as simple as this:
;WITH cte_Values AS
(
SELECT
D.Id
,MAX(D.Value) as [Value]
,ROW_NUMBER() OVER
(
PARTITION BY D.Id
ORDER BY D.DateStamp DESC
) as [Recency]
FROM
SourceTable D
GROUP BY
D.Id
,D.DateStamp
)
-- select out concatenated data, maximum of 30 data points per item
SELECT
C.Id
,dbo.[agg_concat] (CAST(C.Value as nvarchar(30)),',') as [ValueSeries]
FROM
cte_Values C
WHERE
C.Recency <= 30
GROUP BY
C.Id
A nice, simple solution that saves trying to shoehorn some built in SQL function into doing something it was never really intended to do. It just shows that even for DBAs, there are times where it’s much better to use a little managed code to achieve your goal than attempt to force your DBMS of choice to do something at which it’s not so good.
Call it a case of selecting the right tools for the job at hand. Thanks to our resident DB ninja, Ibrahim Naji for finding the path of least resistance.