Goal

There are several reasons causing a node hosting VMs to be not available anymore. Some reasons can be detected by the out-of-band management (or Baseboard Management Controller) of the node and forwarded by the event subscription service. These events should be used to trigger to fencing of the node and to recovery the nodes workload, especially the VMs.

User Stories

• As a cluster admin, I like VMs affected by a crashed node to be rescheduled to another node quickly, so that the downtime of the VMs is below 60 seconds.

Non-Requirements

• Storage based detection and remediation
• Optimize the flow in kubevirt

Notes

• CNV-71846 provides a PoC