An HCP cluster is considered in a state of continuous availability if there is one HCP VM node per KVM host, and if over half of the HCP VM nodes are healthy and running. When the HCP cluster is in this state, the system can survive a single KVM host failure without affecting HCP functionality.
If your HCP cluster is not in a state of continuous availability because you have multiple HCP VM nodes per KVM host and one of your KVM hosts fails, the HCP VM system enters a state of metadata unavailability. Metadata unavailability prohibits HCP namespaces from accepting write requests. The data stored in the affected nodes becomes inaccessible until the HCP system repairs itself. The repair process can take between one and five minutes.
HCP VM systems do not support zero-copy failover. If a namespace has a data protection level of one, the loss of a single HCP VM node causes the node to enter a state of data unavailability until the node is restored.
Oversubscribing the CPU, RAM, or disk of KVM hosts can cause HCP system instability.