In case of eviction (via eviction API) of a VM that can be live migrated, the admission webhook will accept the eviction request marking the VM for eviction (.status.evacuationNodeName).
The eviction controller will try to evict it triggering a live migration. If the migration fails, another one will be retried after a certain time and so on.
If for any reason the VM could not be migrated in a systematically way, the VM will remain in the eviction status forever.

This could block a node drain or external components like the descheduler with a capping on the number of parallel evictions.

4.19

100%

1. force a VM fail a live migration
2. evict it via eviction API
3.

The live migration will be attempted forever

The evict request must eventually come to a conclusion (hard stop, pause ???) within a reasonable but certain time frame (or number of attempts)

We have also have an in progress PR to let the cluster admin cancel an eviction: https://github.com/kubevirt/kubevirt/pull/14587