talos-arc-kvm-unraid

Appearance

Autoscaler Operations ​

Component ​

  • Namespace: arc-systems
  • CronJob: kvm-pool-autoscaler
  • Script: k8s/kvm-pool-autoscaler/scale.sh

Runtime knobs ​

  • RUNNERS_PER_NODE
  • MIN_KVM_NODES
  • MAX_KVM_NODES
  • KVM_NODE_NAMES
  • UNRAID_SSH_TARGET

How it decides scale ​

text
desired = ceil(pending_or_running_dind_pods / RUNNERS_PER_NODE)
desired = clamp(desired, MIN_KVM_NODES, MAX_KVM_NODES)

Operational checks ​

  • Confirm CronJob is scheduled and successful.
  • Inspect logs for virsh start/virsh shutdown behavior.
  • Verify VM power state on libvirt host.

Safe tuning pattern ​

  1. Start with low RUNNERS_PER_NODE for conservative scaling.
  2. Observe queue wait and VM churn.
  3. Increase ratio if churn is too high.
  4. Keep MAX_KVM_NODES aligned with physical capacity.

Failure containment ​

  • If SSH auth fails, autoscaler should fail without mutating VM state.
  • If pod listing fails, job exits and retries on next schedule.