There are plenty of good reasons to move to a virtual infrastructure, and the ability to set up a failover cluster for those virtual servers is key among them. But getting that server set up is not necessarily a straightforward process.
A client of Credera, a management consulting, user experience and technology solutions firm, found that out the hard way when an installation that initially appeared to be problem free failed for reasons that were unclear. The client had two native hypervisor servers they initially attempted to set up in house. “Most folks, when they create a clustered disk, they won’t use a disk they already had,” said Paul Bell, infrastructure architect for Credera and leader on the project. His client, however, set up the installation so that all the virtual machines were stored on a single iSCSI target drive.
The setup passed all of the validations and checks needed to create a cluster, but still faced problems. Because of the shared drive, all of the virtual machines would failover together. The solution was not clear to the client, so Credera was contacted for assistance.
“They basically had the rug pulled out from underneath them whenever the cluster failed,” Bell said. What the client didn’t understand was that—despite following the basic procedure and checks—the installation was flawed.
Examining the installation made it clear to Bell that with the existing storage setup, it wasn’t possible to load balance between the two nodes. What was needed was individual failover roles for the two virtual machines, allowing them to run on separate hosts.
Fixing the problem required setting up a new iSCSI disk as a cluster shared volume (CSV), which meant configuring the disk in one node without assigning it a letter, then using software designed to manage failover clustering to add and identify the new disk. On the new CSV disk, Bell created a subdirectory for each of the virtual machines, then migrated storage to each of them—the longest part of the process.
After the migration was complete, Bell removed the role that contained the two virtual machines, then used a wizard that added new site roles to existing site systems to create a new role for the second virtual machine. For additional virtual machines, that step was just repeated, allowing for load balancing. In this manner, Bell was able to migrate the storage with Hyper-V in Windows Server 2012 on a live Exchange 2010 server without any registered loss of connectivity.
“The process itself is pretty straightforward, assuming you had the experience and background to know what the problem with the cluster installation was and how to solve it,” Bell said. Despite that, he couldn’t find a blog or article with information that clearly outlined the process. So, to try and help others, he wrote one himself.
“The typical how-to article or blog doesn’t take time to dive into details and ‘why’ things are done. Background and experience, benefit and support deploying the right solution—versus deploying a solution right,” Bell said.
The best solution to avoiding Credera’s client’s problem? “Doing it right the first time,” Bell laughed. “But that’s not always the easiest thing to do.” This was one of those instances where the correct path wasn’t completely clear. “They thought everything was fine until they tried their first test failover,” he said—this experience highlights the importance of testing.
“It’s also important to know if the advice you’re getting is the right fit for your particular needs”, Bell said. The blogs and resources you might find are not always correct for your own situation or scenario. This is where a solid background really comes in handy. “Your personal familiarity with the technology can make all the difference,” he said.
Terri Coles is a freelance writer based in St. John’s, NL. Her work covers topics as diverse as food, health and business. If you have a story you would like profiled, contact her at [email protected].
The IT Innovators series of articles is underwritten by Microsoft, and is editorially independent.