Why you should not cluster the virtual appliance
Question
Can I or should I cluster two virtual appliances?
Answer
While it is technically possible to cluster two virtual appliances, it is counterproductive in most cases, and we do not support it. The only use case for clustering our appliance is to avoid hardware outages. In a virtualized environment, the hypervisor should take care of that.
There are several reasons not to cluster two virtual appliances:
- It makes no sense to do clustering on top of virtualization, as the virtualization provides the necessary hardware fault tolerance. For example, vSphere provides two features, depending on your needs: HA or FT. Other solutions have similar approaches.
- When creating snapshots of the appliance, the OS is frozen briefly, which can trigger the failover of the cluster, which in turn will leave you in a split-brain situation. The split-brain happens, due to the fact, that the frozen node will unfreeze shortly thereafter, while still considering itself the primary. As the secondary node took over, one now has two primary nodes, commonly known as a split-brain.
- Network issues on the hypervisor level can affect the cluster in ways that are not obvious. Hence, impacting the cluster and making troubleshooting difficult.
- Generally speaking, you add a level of complexity into an already complex environment, which, in general, you want to avoid.
For the virtual appliance, the clustering feature is only there to use the virtual appliance as a secondary node with a physical appliance as the primary. In this scenario, the virtual appliance will take over only in case of hardware failure of the primary node. As soon as the hardware is up and running, you should make it the primary again.
Related articles