Select non-empty columns using SQL Server - sql

Question

I am using SQL Server 2012. i have a table with 90 columns. I am trying to select only columns that contains data. After searching i used the following procedure:
1- Getting all columns count using one select query
2- Pivoting Result Table into a Temp table
3- Creating Select query
4- Executing this query
Here is the query i used:
DECLARE #strTablename varchar(100) = 'dbo.MyTable'
DECLARE #strQuery varchar(max) = ''
DECLARE #strSecondQuery varchar(max) = 'SELECT '
DECLARE #strUnPivot as varchar(max) = ' UNPIVOT ([Count] for [Column] IN ('
CREATE TABLE ##tblTemp([Column] varchar(50), [Count] Int)
SELECT #strQuery = ISNULL(#strQuery,'') + 'Count([' + name + ']) as [' + name + '] ,' from sys.columns where object_id = object_id(#strTablename) and is_nullable = 1
SELECT #strUnPivot = ISNULL(#strUnPivot,'') + '[' + name + '] ,' from sys.columns where object_id = object_id(#strTablename) and is_nullable = 1
SET #strQuery = 'SELECT [Column],[Count] FROM ( SELECT ' + SUBSTRING(#strQuery,1,LEN(#strQuery) - 1) + ' FROM ' + #strTablename + ') AS p ' + SUBSTRING(#strUnPivot,1,LEN(#strUnPivot) - 1) + ')) AS unpvt '
INSERT INTO ##tblTemp EXEC (#strQuery)
SELECT #strSecondQuery = #strSecondQuery + '[' + [Column] + '],' from ##tblTemp WHERE [Count] > 0
DROP TABLE ##tblTemp
SET #strSecondQuery = SUBSTRING(#strSecondQuery,1,LEN(#strSecondQuery) - 1) + ' FROM ' + #strTablename
EXEC (#strSecondQuery)
The problem is that this query is TOO SLOW. Is there a best way to achieve this?
Notes:
Table have only one clustered index on primary key Column ID and does not contains any other indexes.
Table is not editable.
Table contains very large data.
Query is taking about 1 minute to be executed
Thanks in advance.

Answer 1

I do not know if this is faster, but you might use one trick: FOR XML AUTO will ommit columns without content:
DECLARE #tbl TABLE(col1 INT,col2 INT,col3 INT);
INSERT INTO #tbl VALUES (1,2,NULL),(1,NULL,NULL),(NULL,NULL,NULL);
SELECT *
FROM #tbl AS tbl
FOR XML AUTO
This is the result: col3 is missing...
<tbl col1="1" col2="2" />
<tbl col1="1" />
<tbl />
Knowing this, you could find the list of columns, which are not NULL in all rows, like this:
DECLARE #ColList VARCHAR(MAX)=
STUFF
(
(
SELECT DISTINCT ',' + Attr.value('local-name(.)','nvarchar(max)')
FROM
(
SELECT
(
SELECT *
FROM #tbl AS tbl
FOR XML AUTO,TYPE
) AS TheXML
) AS t
CROSS APPLY t.TheXML.nodes('/tbl/#*') AS A(Attr)
FOR XML PATH('')
),1,1,''
);
SELECT #ColList
The content of #ColList is now col1,col2. This string you can place in a dynamically created SELECT.
UPDATE: Hints
It would be very clever, to replace the SELECT * with a column list created from INFORMATION_SCHEMA.COLUMNS excluding all not-nullable. And - if needed and possible - types, wich contain very large data (BLOBs).
UPDATE2: Performance
Don't know what your very large data means actually... Just tried this on a table with about 500.000 rows (with SELECT *) and it returned correctly after less than one minute. Hope, this is fast enough...

Answer 2

Try using this condition:
where #columnname IS NOT NULL AND #columnname <> ' '