Archive for the ‘TempDB’ Category

Last week on one of our production servers (version 2011.110.3373.0), we’ve encountered a strange issue which is relatively new (only applicable for SQL Server 2012 and 2014 versions). Based on the symptoms observed and doing a quick Bing search found we encountered a product bug!

I observed multiple SPIDs, around 300+ running the same stored procedure which is doing an INSERT operation and are in a suspended status, they seemed to be hanging there for atleast like 45 minutes (usually they finish in few seconds). There is no blocking and I found nothing wrong with the SQL Server except for started seeing elevated values for PWAIT_MD_RELATION_CACHE and MD_LAZYCACHE_RWLOCK wait types. And, oh! I also noticed that tempdb data was at 99% full.

So I thought I found the root cause and tried mitigating the problem by adding extra tempdb space but still wasn’t any help. As the last resort I had to failover the services to another node, basically restarted SQL Server, to bring the server fully functional.

On further investigation we found the root cause as someone tried to create an index (online) to improve performance, but later he cancelled the create index and we hit this bug. However, this bug only effected the table on which index operation was attempted and so, everything else was operational except for that Stored Procedure running multiple SPIDs trying to INSERT to the table on which the user attempted CREATE INDEX command was blocked, impairing that part of functionality of the application.

Resolution: Cumulative Update 9 for SQL Server 2012 SP1; Cumulative Update 1 for SQL Server 2014

Here are more details about this problem: http://support.microsoft.com/kb/2926712


Read Full Post »

In order to reduce Tempdb contention one of  the best practices is to maintain multiple sized Tempdb data files, matching the number of processors and up to a maximum of 8. In this post I will show you T-SQL script to identify current Tempdb configuration and number of logical processors along with adding additional Tempdb data files as required.

Script 1: Find current tempdb configuration

select DB_NAME(mf.database_id) database_name
, mf.name logical_name, mf.file_id
, (CONVERT(DECIMAL, size)/128)) as [file_size_MB]
, CASE mf.is_percent_growth
END AS [is_percent_growth]
, CASE mf.is_percent_growth
WHEN 1 THEN CONVERT(VARCHAR, mf.growth) + '%'
WHEN 0 THEN CONVERT(VARCHAR, mf.growth/128) + ' MB'
END AS [growth_in_increment_of]
, CASE mf.is_percent_growth
,(((CONVERT(DECIMAL, size)*growth)/100)*8)/1024)
, (CONVERT(DECIMAL, growth)/128))
END AS [next_auto_growth_size_MB]
, physical_name from sys.master_files mf
where database_id =2 and type_desc= 'rows'

Script 2: Find number of logical processors

SELECT cpu_count AS logicalCPUs FROM sys.dm_os_sys_info

Script 3: Add tempdb data files as per processor count from the above query

ALTER DATABASE tempdb ADD FILE ( NAME = N'tempdev2',
FILENAME = N'D:\DBA\Data\tempdev2.ndf' , SIZE =8MB , FILEGROWTH = 5MB) --<<--Update the data file location/Size/AutoGrowth

ALTER DATABASE tempdb ADD FILE ( NAME = N'tempdev3',
FILENAME = N'D:\DBA\Data\tempdev3.ndf' , SIZE =8MB , FILEGROWTH = 5MB)--<<--Update the data file location/Size/AutoGrowth
---ETC, add files as per processors count

Reboot/Restart of SQL services is not required for making the tempdb changes. Here is a great post that explains how to best remove extra Tempdb files.

For more information:


Read Full Post »

One of the Microsoft’s recommendation for optimizing the tempDB performance is to make each tempdb data file the same size.

Today, on one of our servers, I noticed there are 13 data files with different sizes as shown in the below screenshot:


My target here is to configure tempdb with 8 equi sized data files and one log file. So, in this case I have to delete those 5 extra data files and re-size the remaining 8 files equally.

To achieve this I followed the below simple three step process (The same procedure can be applied to any user databases as well)

-- Step1: First empty the data file
USE tempdb
DBCC SHRINKFILE (tempdev12, EMPTYFILE); -- to empty "tempdev12" data file

Data file won’t gets deleted unless it is empty. So before going to Step2 to delete the data file, we should first empty the data file which can be done from the Step1 above.

Note: If encountered this error Msg 2555, Level 16, State 1, Line 1 while performing the above operation, please refer to this post –> Msg 2555, Level 16, State 1, Line 1 – Error DBCC SHRINKFILE with EMPTYFILE option

--Step2: Remove that extra data file from the database
REMOVE FILE tempdev12; --to delete "tempdev12" data file

I repeated the above two steps to delete the other files as well. Below is the screenshot after deleting the extra files


--Step3: Re-size the data files to target file size 
-- Use ALTER DATABASE if the target file size is greater than the current file size
USE [master]
MODIFY FILE ( NAME = N'tempdev', SIZE = 3072000KB ) --grow to 3000 MB
--Use DBCC SHRINKFILE if the target file size is less than the current file size
USE [tempdb]
DBCC SHRINKFILE (N'tempdev' , 3000) --shrink to 3000 MB

In my case, the target file size is 3000 MB which is greater than the current file size, so I ran the ALTER DATABASE command for all the 8 files. And below is the screenshot after re-sizing the files


Note: The same procedure can be used for removing extra log files as well. Make sure the log file you are trying to empty do not have any active portion of the log. Run DBCC LOGINFO on your specified database, taking note of the FileId and the Status indicator (2 showing that it is an active VLF)

Read Full Post »