SBD (Storage based death) is a fencing mechanism to halt nodes that are in inconsistent state. In order to be used all the clustered nodes must have a shared disk.
Find more information in:
The next options are available to use SBD as our cluster fencing mechanism.
Use a shared disk served by an ISCSI server. This is a quite standard option if the clusters hosted in the cloud, as shared disks are not commonly available. To use this option, an ISCSI server must be created (or use an already existing one). The project gives the option to create a new virtual machine to host this service. For that we need to use the next variables:
- Set
sbd_storage_typetoiscsi - Enable at least one cluster that will use SBD setting the option
*_cluster_sbd_enabledtotrue - ISCSI server configuration has some advanced configuration options. Check the terraform template examples and the available variables for that.
Use a shared disk attached to all clustered nodes. To use this option:
- Set
sbd_storage_typetoshared-disk
The usage of SBD using ISCSI is mandatory by now in Azure, as it doesn't provide any other fencing mechanism. AWS and GCP have native fencing mechanisms, so SBD usage is optional. A deployment done in Libvirt can use ISCSI or shared disk, both are available, but one of them must be provided.