Today’s post is a guest article from a friend of Dallas DBAs, writer, and fantastic DBA Jules Behrens (B|L)
One common performance issue that is not well known that should still be on your radar as a DBA is a high number of VLFs.
Virtual Log Files are the files SQL Server uses to do the actual work in a SQL log file (MyDatabase_log.LDF). It allocates new VLFs every time the log file grows. Perhaps you’ve already spotted the problem – if the log file is set to grow by a tiny increment, then if your the file ever grows very large, you may end up with thousands of tiny little VLFs, and this can slow down your performance at the database level. Think of it like a room (the log file) filled with boxes (the VLFs). If you just have a few boxes, it is more efficient to figure out where something (a piece of data in the log file) is, than if you have thousands of tiny boxes. (Analogy courtesy of @SQLDork)
It is especially evident there is an issue with VLFs when SQL Server takes a long time to recover from a restart. Other symptoms may be slowness with autogrowth, log shipping, replication, and general transactional slowness. Anything that touches the log file, in other words.
The best solution is prevention – set your log file to be big enough to handle its transaction load to begin with, and set it to have a sensible growth rate in proportion to its size, and you’ll never see this come up. But sometimes we inherit issues where best practices were not followed, and a high number of VLFs is certainly something to check when doing a health assessment on an unfamiliar environment.
Find your VLF counts
The built-in DMV sys.dm_db_log_info is specifically for finding information about the log file, and command DBCC LOGINFO (deprecated) will return a lot of useful information about VLFs as well. There is an excellent script for pulling the count of VLFs that uses DBCC LOGINFO from Kev Riley, on Microsoft Tech Net:
https://gallery.technet.microsoft.com/scriptcenter/SQL-Script-to-list-VLF-e6315249
There is also a great script by Steve Rezhener on SQLSolutionsGroup.com that utilizes the view:
https://sqlsolutionsgroup.com/capture-sql-server-vlf-information-using-a-dmv/
Either one of these will tell you what you ultimately need to know – if your VLFs are an issue. How many VLFs are too many? There isn’t an industry standard, but for the sake of a starting point, let’s say a tiny log file has 500 VLFs. That is high. A 5GB log file with 200 VLFs, on the other hand, is perfectly acceptable. You’ll likely know a VLF problem when you find it; you’ll run a count on the VLFs and it will return something atrocious like 20,000. (ed – someone at Microsoft support told me about one with 1,000,000 VLFs)
Resolution
If the database is in Simple recovery model and doesn’t see much traffic, this is easy enough to fix. Manually shrink the log file as small as it will go, verify the autogrow is appropriate, and grow it back to its normal size. If the database is in Full recovery model and is in high use, it’s a little more complex. Follow these steps (you may have to do it more than once):
- Take a transaction log backup .
- Issue a CHECKPOINT manually.
- Check the empty space in the transaction log to make sure you have room to shrink it.
- Shrink the log file as small as it will go.
- Grow the file back to its normal size.
- Lather, Rinse, Repeat as needed
Now check your VLF counts again, and make sure you are down to a nice low number. Done!
If you need help, contact us
Thanks for reading!
Kevin3NF