Like most SQL Server users I'm rather frustrated by Microsoft's insistence on making the really cool features only available in Enterprise Edition. And it really doesn't help that they changed the licensing for SQL 2012 to be core-based, so now it's like 4 times as expensive! It almost makes you want to go with Oracle. That, and a desire to have Larry Ellison do things to your orifices.
And since they've introduced Availability Groups, and marked database mirroring as deprecated, you'd think they'd make make mirroring available in all editions. Alas…they don't…officially anyway. Thanks to my constant poking around in places I'm not "supposed" to, I've discovered the low-level code that implements database mirroring, and found that it's available in all editions!
It turns out that the query processor in all SQL Server editions prepends a simple check before every edition-specific DDL statement:
IF CAST(SERVERPROPERTY('Edition') as nvarchar(max)) NOT LIKE '%e%e%e% Edition%'
If that statement returns true, it fails. (the print statements are just placeholders) Go ahead and test it on Standard, Workgroup, and Express editions compared to an Enterprise or Developer edition instance (which support everything).
Once again thanks to Argenis Fernandez (b | t) and his awesome sessions on using Sysinternals, I was able to watch the exact process SQL Server performs when setting up a mirror. Surprisingly, it's not actually implemented in SQL Server! Some of it is, but that's something of a smokescreen, the real meat of it is simple filesystem primitives.
The NTFS filesystem supports links, both hard links and symbolic, so that you can create two entries for the same file in different directories and/or different names. You can create them using the MKLINK command in a command prompt:
mklink /D D:\SkyDrive\Data D:\Data
mklink /D D:\SkyDrive\Log D:\Log
This creates a symbolic link from my data and log folders to my Skydrive folder. Any file saved in either location will instantly appear in the other. And since my Skydrive will be automatically synchronized with the cloud, any changes I make will be copied instantly (depending on my internet bandwidth of course).
So what does this have to do with database mirroring? Well, it seems that the mirroring endpoint that you have to create between mirror and principal servers is really nothing more than a Skydrive link. Although it doesn't actually use Skydrive, it performs the same function. So in effect, the following statement:
ALTER DATABASE Mir SET PARTNER='TCP://MyOtherServer.domain.com:5022'
Is turned into:
mklink /D "D:\Data" "\\MyOtherServer.domain.com\5022$"
The 5022$ "port" is actually a hidden system directory on the principal and mirror servers. I haven't quite figured out how the log files are included in this, or why you have to SET PARTNER on both principal and mirror servers, except maybe that mklink has to do something special when linking across servers. I couldn't get the above statement to work correctly, but found that doing mklink to a local Skydrive folder gave me similar functionality.
To wrap this up, all you have to do is the following:
- Install Skydrive on both SQL Servers (principal and mirror) and set the local Skydrive folder (D:\SkyDrive in these examples)
- On the principal server, run mklink /D on the data and log folders to point to SkyDrive: mklink /D D:\SkyDrive\Data D:\Data
- On the mirror server, run the complementary linking: mklink /D D:\Data D:\SkyDrive\Data
- Create your database and make sure the files map to the principal data and log folders (D:\Data and D:\Log)
- Viola! Your databases are kept in sync on multiple servers!
One wrinkle you will encounter is that the mirror server will show the data and log files, but you won't be able to attach them to the mirror SQL instance while they are attached to the principal. I think this is a bug in the Skydrive, but as it turns out that's fine: you can't access a mirror while it's hosted on the principal either. So you don't quite get automatic failover, but you can attach the files to the mirror if the principal goes offline. It's also not exactly synchronous, but it's better than nothing, and easier than either replication or log shipping with a lot less latency.
I will end this with the obvious "not supported by Microsoft" and "Don't do this in production without an updated resume" spiel that you should by now assume with every one of my blog posts, especially considering the date.