With KubeVirt server v1.8.0-beta.0-331-g8711eeaca0 (and virtctl client 1.8.0), the serial console connection becomes unusable when a long-running process runs inside the VM (e.g. a container with a long or heavy workload)

OCP 4.22.0-ec.2
CNV 4.22.0-17
Client: v1.8.0-beta.0 (or matching 1.8.x)
Server: v1.8.0-beta.0-331-g8711eeaca0

Create a VM that boots and has cloud-init run a long-running workload (e.g. podman run ... that runs for several minutes and prints to stdout), or log into the VM and start such a command manually.

runcmd:
  - echo "Flooding serial..."
  - bash -c 'for i in $(seq 1 50000); do echo "line $i"; done'
  - echo "Done"
Connect to the VM serial console with virtctl:
virtctl console -n <namespace> <vm_name>
and/or capture it:
virtctl console -n <namespace> <vm_name> > /tmp/vm.log. 

Observe: the console disconnects with 1006 / unexpected EOF, and/or the capture file stays empty or stops updating (console appears stuck), while the workload inside the VM continues and produces output when accessed interactivel

VM should finish the run w/o any error

- Works on OCP 4.21 with KubeVirt 1.7.1 (same VM image and workload).
- Fails on OCP 4.22 with KubeVirt 1.8.x.
- Both client and server are 1.8.x (no version skew). virtctl is used to capture the serial console for CI (benchmark-runner); the VM runs a container that can run for 10+ minutes with continuous serial output.
Reducing kernel log level in the VM and redirecting container stderr to a file reduce serial noise but do not fix the disconnect or stuck capture.