Skip to content

Power Toggle VMs

Power VMs on or off in bulk, with --action {on, off} selecting the direction. Designed to be symmetric: a single power-off run produces a snapshot file that a subsequent power-on run can consume to restore the exact same set of VMs.

Use Case: Free up node capacity before a benchmark, simulate partial node failure, or repeatedly power-cycle subsets of a fleet.

How It Works

--action off

  1. Lists all running VMIs on --node.
  2. Randomly samples --percentage of them.
  3. Saves the selection to powered_off_vms_<node>_<YYYYMMDD_HHMMSS>.txt (one namespace/name per line).
  4. Issues virtctl stop (or kubectl virt stop) to every selected VM in parallel.
  5. Waits up to --wait-timeout for each VMI to disappear or reach Succeeded/Failed.

--action on

Powered-off VMs do not have a nodeName on their VMI (they have no VMI at all), so node-only discovery is not possible. You must specify a namespace range or a list file:

  • Range mode — iterate <namespace-prefix>-<i> for i in [start, end], looking for --vm-name in each namespace. Optionally filter by --node (matches the VM's spec.template.spec.nodeSelector.kubernetes.io/hostname).
  • List-file mode — read namespace/name lines from --vm-list-file (typically the file produced by a previous --action off run).

The command then issues virtctl start to every target in parallel and waits for each VMI to reach Running.

Basic Usage

Using virtbench CLI

# Power off 50% of running VMs on a node
virtbench vm-ops power-toggle-vms \
  --action off --node worker-1 --percentage 50

# Power them back on from the saved list
virtbench vm-ops power-toggle-vms \
  --action on --vm-list-file powered_off_vms_worker-1_20260508_120000.txt

# Power on a namespace range (no list file required)
virtbench vm-ops power-toggle-vms \
  --action on \
  --namespace-prefix migration --start 1 --end 50 \
  --vm-name rhel-9-vm

# Power on only those VMs whose nodeSelector pins them to worker-1
virtbench vm-ops power-toggle-vms \
  --action on \
  --namespace-prefix migration --start 1 --end 50 \
  --vm-name rhel-9-vm \
  --node worker-1

Full Example

# Aggressive power-off: 75% of node, higher concurrency
virtbench vm-ops power-toggle-vms \
  --action off \
  --node worker-2 \
  --percentage 75 \
  --concurrency 100 \
  --wait-timeout 600

# Symmetric power-on from the same list, dry-run first
virtbench vm-ops power-toggle-vms \
  --action on \
  --vm-list-file powered_off_vms_worker-2_20260508_120000.txt \
  --concurrency 100 \
  --wait-timeout 600 \
  --dry-run

Options

Option Description
--action (required) on or off.
--node Required for --action off (unless --vm-list-file). Optional nodeSelector filter for --action on.
--percentage Percentage of running VMs to power off (default: 50, --action off only).
--namespace-prefix, --start, --end, --vm-name Range-based discovery for --action on.
--vm-list-file File of namespace/name lines to act on (works with both actions).
--concurrency Max concurrent operations (default: 50).
--wait-timeout Timeout waiting for VMs to reach the target phase (default: 300s).
--dry-run Show what would be done without doing it.

Output

The summary at the end of each run reports:

  • VMs targeted
  • start / stop commands sent vs. failed
  • VMs that reached the target phase (Running for on, Succeeded/ Failed/missing for off)
  • Wall-clock breakdown (command-send time vs. wait time vs. total)

For --action off, the snapshot file path is also logged so you can feed it back into --action on later.

Notes

  • virtctl (or the kubectl virt Krew plugin) must be on $PATH. The script tries the Krew plugin first, then the standalone binary.
  • --vm-list-file lines look like namespace/name; blank lines and lines without a / are ignored.
  • --action off always saves a snapshot file before issuing any stop commands. If the run is interrupted, you can still resume with --action on --vm-list-file <that file>.

See Also