(The Curious Case of… used to be part of our bi-weekly newsletter but we decided to make it a regular blog post instead so it can sometimes be more frequent. It covers something interesting one of us encountered when working with a client, doing some testing, or were asked in a random question from the community.)
I was working recently on a client’s corruption issue where I needed to know how to find which page an index row is on. The client was getting weird errors from DBCC CHECKDB around missing nonclustered index records, but SELECTs using that index on the supposedly missing key values would work. I suspected a stale read problem from the I/O subsystem as part of building the database snapshot that DBCC CHECKDB uses, but needed to prove which page the issue was on.
There’s a cool undocumented column that you can select called %%PHYSLOC%% that gives the page:record:slot of the selected record, and I’ve used it many times when working on corruption/data recovery issues for clients. It gives the information back in hex, so there’s a companion function called sys.fn_PhysLocCracker that formats the output nicely. Both of these have been in the product since SQL Server 2005.
Let me show you an example using the old AdventureWorks sample database.
Let’s say there’s a corruption issue where it says there’s a missing row for StateProvinceID = 1 and AddressID = 519 in the IX_Address_StateProvinceID nonclustered index (which is index ID 4) of the Person.Address table. If I want to prove that that row is NOT missing, here’s what I can do:
SELECT [StateProvinceID] , [AddressID] , [physloc].* FROM [Person].[Address] WITH (INDEX = 4) CROSS APPLY sys.fn_PhysLocCracker (%%physloc%%) AS [physloc] WHERE [StateProvinceID] = 1 AND [AddressID] = 519; GO
I’m selecting the index keys for the nonclustered index and using an index hint to force that index to be used.
And if I get any output, the row exists. In this case, I get:
StateProvinceID AddressID file_id page_id slot_id --------------- ----------- ----------- ----------- ----------- 1 519 1 8120 9
I’ve also used this to find rows that I need to bit-twiddle using DBCC WRITEPAGE to work around corruptions to allow data recovery from trashed databases – tedious, but possible if you know what you’re doing.
Bottom line: there are quite a few little undocumented columns and functions you can use to easily get internals information about SQL Server, with real-world applications.