Resizing Tempdb (When TEMPDB Wont Shrink)

Occasionally, we must resize or realign our Tempdb log file (.ldf) or data files (.mdf or .ndf) due to a growth event that forces the file size out of whack. To resize we have three options, restart the SQL Server Service, add additional files, or shrink the current file. We most likely have all been faced with run away log files and in an emergency situation restarting the SQL Services may not be an option but we still need to get our log file size smaller before we run out of disk space for example. The process of shrinking down that file can get tricky so I created this flow chart to help you out if you ever get into this situation.

Now its very important to note that many of these commands will clear your cache and will greatly impact your server performance as it warms cache backup. In addition, you should not shrink your database data or log file unless absolutely necessary. But doing so,  it can result in a corrupt tempdb.

Let’s walk through it and explain somethings as we go along.

First thing you must do is issue a Checkpoint. A checkpoint marks the log as a “good up to here” point of reference. It lets the SQL Server Database Engine know it can start applying changes contained in the log during recovery after this point if an unexpected shutdown or crash occurs. Anything prior to the check point is what I like to call “Hardened”. This means all the dirty pages in memory have been written to disk, specifically to the .mdf and .ndf files. So, it is important to make that mark in the log before you proceed. Now, we know tempdb is not recovered during a restart it is recreated, however this is still a requirement.

USE TEMPDB;  
GO  
CHECKPOINT;

Next, we try to shrink the log by issuing a DBCC SHRINKFILE command. This is the step that frees the unallocated space from the database file if there is any unallocated space available. You will note the Shrink? decision block in the diagram after this step. It is possible that there is no unallocated space and you will need to move further along the path to free some up and try again.

USE TEMPDB;  
GO 
DBCC SHRINKFILE (templog, 1000);   --Shrinks it to 1GB

If the database shrinks, great congratulations, however for some of us we still might have work to do. Next up is to try and free up some of that allocated space by running DBCC DROPCLEANBUFFERS and DBCC FREEPROCCACHE.

DBCC DROPCLEANBUFFERS – Clears the clean buffers from the buffer pool and columnstore object pool. This will flush cached indexes and data pages.

DBCC DROPCLEANBUFFERS WITH NO_INFOMSGS;

DBCC FREEPROCCACHE – Clears the procedure cache, you are probably familiar with as a performance tuning tool in development. It will clean out all your execution plans from cache which may free up some space in tempdb. As we know though, this will create a performance  as your execution plans now have to make it back into cache on their next execution and will not get the benefit of plan reuse.  Now it’s not really clear why this works, so I asked tempdb expert Pam Lahoud (B|T) for clarification as to why this has anything to do with tempdb. Both of us are diving into this to understand exactly why this works. I believe it to be related to Tempdb using cached objects and memory objects associated with stored procedures which can have latches and locks on them that need to be release by running this. Check back for further clarification, I’ll be updating this as I find out more.

DBCC FREEPROCCACHE WITH NO_INFOMSGS;

Once these two commands have been run and you have attempted to free up some space you can now try the DBCC SHRINKFILE command again. For most this should make the shrink possible and you will be good to go.  Unfortunately, a few more of us may have to take a couple more steps through to get to that point.

The last two things I do when I have no other choice to get my log file smaller is to run those last two commands in the process. These should do the trick and get the log to shrink.

DBCC FREESESSIONCACHE– This command will flush any distributed query connection cache, meaning queries that are between two or more servers.

DBCC FREESESSIONCACHE WITH NO_INFOMSGS;

DBCC FREESYSTEMCACHE – This command will release all unused remaining cache entries from all cache stores including temp table cache. This covers any temp table or table variables remaining in cache that need to be released.

DBCC FREESYSTEMCACHE ('ALL');

In my early days as a database administrator I would have loved to have this diagram. Having some quick steps during stressful situations such as tempdb’s log file filling up on me would have been a huge help. So hopefully someone will find this handy and will be able to use it to take away a little of their stress.

Share

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Trust DCAC with your data

Your data systems may be treading water today, but are they prepared for the next phase of your business growth?