Recently, I was working on one of the performance tuning assignment with an esteemed client. The size of their databases were growing tremendously. Growth in the database size is directly proportionate to disk space requirement. More the database size will be, more disk space will be required.
When we talk about the disk space for Production, it’s not only specific to one server. There are multiple factors that needs to be considered such as Primary server, Secondary servers, Backup servers and Backups (Full, Differential, Log backups).
Long time ago, I blogged about similar topic. Read the full article here.
While we were discussing the various possibilities to reduce the database size, client brought up a very interesting point in front of me. He said “My database has many huge tables (with size beyond 500 GB’s). Most of them have multiple columns which has null values across the table.”
Although, client was aware that there are many such tables and columns but there was not definite list that client has. This has opened a whole new arena for me and gave a unique insights.
I started thinking to build some logic to fetch the list of all such columns and tables. You will be amazed to hear but it’s true. There were more than 4K such columns belonging to 500+ tables. We estimated that we could save TB’s of disk space by marking all such columns as SPARSE.
More on SPARSE columns can be read here at Microsoft official documentation.
In this article we’ll see how to get the list of columns which has null values across all the rows of the table.
Note: Running the query for all the tables together would take lot of time. It is recommended to run the query for set of tables in batches parallely in multiple instances of SSMS or with the help of jobs.
In the next article, we’ll talk about how to estimate the storage savings by removing such columns or marking them as SPARSE.
IF NOT EXISTS (SELECT 1 FROM sys.tables WHERE name = 'tables_with_null_values_across')
BEGIN
CREATE TABLE tables_with_null_values_across
(
TableName VARCHAR(200)
, TotalRows BIGINT
, ColumnName VARCHAR(200)
)
END
/*
DROP TABLE IF EXISTS #nullable_columns;
DROP TABLE IF EXISTS #tables_with_nullable_columns;
*/
SET NOCOUNT ON
SELECT A.object_id
, B.name AS TableName
, A.name AS ColumnName
, ROW_NUMBER() OVER(PARTITION BY A.object_id ORDER BY A.name ASC) AS RowID
INTO #nullable_columns
FROM sys.columns A
INNER JOIN sys.tables B
ON B.object_id = A.object_id
AND B.type = 'U'
LEFT JOIN tables_with_null_values_across D
ON D.TableName = QUOTENAME(B.name)
AND D.ColumnName = QUOTENAME(A.Name)
WHERE A.is_nullable = 1
AND D.TableName IS NULL
-- AND B.name IN ('', '')
-- Note: Supply the table names in the filter clause for B.name in order to run the query in batches of tables.
SELECT DISTINCT A.object_id
, B.name AS TableName
, IDENTITY(INT, 1, 1) AS RowID
INTO #tables_with_nullable_columns
FROM #nullable_columns A
INNER JOIN sys.tables B
ON B.object_id = A.object_id
DECLARE @TableName AS SYSNAME
, @ColumnName AS SYSNAME
, @Object_ID AS INT
, @Table_RowID AS INT
, @Column_RowID AS INT
, @Column_Distinct_Values_Count AS BIGINT
, @Total_Rows AS BIGINT
SET @Table_RowID = 1;
WHILE EXISTS (SELECT 1 FROM #tables_with_nullable_columns WHERE RowID = @Table_RowID)
BEGIN
SELECT @Object_ID = object_id
, @TableName = TableName
FROM #tables_with_nullable_columns
WHERE RowID = @Table_RowID;
SET @Column_RowID = 1;
WHILE EXISTS (SELECT 1 FROM #nullable_columns WHERE object_id = @Object_ID AND RowID = @Column_RowID)
BEGIN
SELECT @ColumnName = ColumnName
FROM #nullable_columns
WHERE object_id = @Object_ID
AND RowID = @Column_RowID;
DECLARE @SQLString NVARCHAR(500);
SET @SQLString = N'SELECT @Column_Distinct_Values_Count = COUNT(DISTINCT ' + QUOTENAME(@ColumnName) + ')
, @Total_Rows = COUNT(1)
FROM ' + QUOTENAME(@TableName) + ' WITH (NOLOCK)';
BEGIN TRY
EXECUTE sp_executesql @SQLString
, N'@Total_Rows BIGINT OUTPUT, @Column_Distinct_Values_Count BIGINT OUTPUT'
, @Total_Rows = @Total_Rows OUTPUT
, @Column_Distinct_Values_Count = @Column_Distinct_Values_Count OUTPUT ;
END TRY
BEGIN CATCH
END CATCH
IF (@Column_Distinct_Values_Count = 0)
BEGIN
INSERT INTO tables_with_null_values_across (TableName, TotalRows, ColumnName)
VALUES (QUOTENAME(@TableName), @Total_Rows, QUOTENAME(@ColumnName))
END
SET @Column_RowID = @Column_RowID + 1;
END
SET @Table_RowID = @Table_RowID + 1;
END
DROP TABLE IF EXISTS #nullable_columns;
DROP TABLE IF EXISTS #tables_with_nullable_columns;
I think i missed a post. Where does `tables_with_same_values_across` come from?
LikeLike
Hi Peter,
You didn’t missed any post. Actually it was a miss from my side. Actually I should have removed the INNER JOIN table_with_same_values_across from the query.
This same query with slight modification can be used to fetch the list of columns having same values across. In my assignment, I first extracted the list of columns with same values across the tables of a database and then out of this list extracted the columns with null values.
Thanks for pointing out the error. I’ve made the necessary corrections and also will soon write a new post for finding the columns with same values across the table.
LikeLike
Now there is no reference for fields C.TableName and C.ColumnName ?
LikeLike
Wise people say “over smartness kills”. This instance is real life example 🙂
I made the changes directly in the post without even running it.
I ran the query this time, validated the result and then modified the post.
Apologies for the inconvenience caused and Thanks for your time and efforts for pointing out the errors!
LikeLike
Does the client have SQL data compression available? If so, are they using it? If not, I’d definitely focus on that first as it should have an order of magnitude more overall size effect than NULL columns.
LikeLike
Not yet. But, we are in the process of implementing the table compression too.
Quite often such kinds of small points having huge impacts are ignored. It was an eye opener for the client and their Dev. team. How come a database has thousands of columns having only null values across? I mean it seriously doesn’t makes any sense of having such columns. This analysis has made the client to relook at their existing SDLC process so that they can avoid similar kinds of gaps at early stage only.
LikeLike
You should update the code.
LikeLike