Disk Operations Benchmark

Tests disk hotplug and coldplug performance for KubeVirt VMs.

Use Case: Measure disk attach/detach times for running and stopped VMs with in-VM validation.

Operations

Operation	Description	VM State
hotplug	Attach disk to running VM	Running
coldplug	Attach disk to stopped VM, then start	Stopped
all	Run both hotplug and coldplug	Both

Basic Usage

virtbench CLI

# Create VMs and hotplug 3 disks
virtbench disk-ops --start 1 --end 10 --storage-class px-csi --disks 3 \
  --create-vms --save-results

# Coldplug test (creates VMs, stops them, attaches disk, starts them)
virtbench disk-ops --start 1 --end 5 --storage-class px-csi \
  --operation coldplug --create-vms --save-results

# All operations with unplug test and cleanup
virtbench disk-ops --start 1 --end 10 --storage-class px-csi \
  --operation all --create-vms --test-unplug --cleanup

Python Script

cd disk-ops-benchmark

python3 measure-disk-ops.py \
  --start 1 --end 10 \
  --storage-class px-csi \
  --operation hotplug \
  --disks 3 \
  --create-vms \
  --save-results

Validation Modes

In-VM validation runs lsblk over SSH to confirm each disk is visible inside the guest. SSH goes through a persistent helper pod (--ssh-pod, default ssh-test-pod in the default namespace) that has sshpass installed — the pod is auto-created if it doesn't exist and reused across runs. Authentication is password-based (--vm-user / --vm-password).

• --create-vms — VMs are provisioned with --vm-password baked into their cloud-init, so validation works out of the box.
• Pre-existing VMs — pass --vm-user / --vm-password matching the password those VMs already accept (and ensure their sshd allows password auth).
• --skip-validation — skip the in-VM SSH check entirely and rely only on API-level volume status. No helper pod is created.

Configuration Options

Operation Settings

Option	Default	Description
--start	(required)	Start namespace index
--end	(required)	End namespace index
--storage-class	(required)	Storage class for PVCs
--operation	all	Operation: hotplug, coldplug, all
--disks	1	Number of disks per VM
--disk-size	10Gi	Disk size
--test-unplug	false	Also test disk removal

VM Settings

Option	Default	Description
--namespace-prefix	disk-ops	Namespace prefix
--vm-name	disk-ops-vm	VM name
--vm-template	disk-ops-benchmark/vm-template.yaml	VM template file (used with --create-vms)
--vm-user	cloud-user	VM SSH user for in-VM validation
--vm-password	changeme	VM SSH password for in-VM validation
--ssh-pod	ssh-test-pod	Persistent SSH helper pod with sshpass (auto-created if missing)
--ssh-pod-ns	default	SSH helper pod namespace
--create-vms	false	Create VMs before testing

Execution Settings

Option	Default	Description
--concurrency	10	Max concurrent operations
--attach-timeout	300	Disk attach timeout (seconds)
--skip-validation	false	Skip in-VM validation

Output Settings

Option	Default	Description
--results-dir	results	Base results directory
--save-results	false	Save results to JSON/CSV
--px-version	px-unknown	Storage driver/version label for the results folder
--disk-type	1-disk	Disk type label for the results folder
--cleanup	false	Remove hotplugged disks (and created VMs) after test

Metrics

Hotplug Metrics

Metric	Description
api_attach_time	Time for virtctl addvolume API call
volume_ready_time	Time for volume to appear in VMI status
validation_time	Time to verify disk in VM via SSH
total_time	End-to-end hotplug time

Coldplug Metrics

Metric	Description
api_attach_time	Time to attach disk to stopped VM
vm_boot_time	Time for VM to boot with new disk
validation_time	Time to verify disk in VM via SSH
total_time	End-to-end coldplug time

Validation

The benchmark validates disk attachment at two levels:

1. API Level: Checks volume appears in vmi.status.volumeStatus
2. VM Level: SSH into VM and verify disk via lsblk

Validation ensures the disk is actually usable inside the VM, not just attached at the API level.

Results

Output Structure

results/{px-version}/{disk-type}/{timestamp}_disk_ops_benchmark_{start}-{end}/
├── disk_ops_results.json    # Full results with per-VM data
└── disk_ops_summary.csv     # CSV summary

The {px-version} and {disk-type} segments come from --px-version and --disk-type, keeping the layout compatible with the results dashboard.

Sample Output

======================================================================
DISK OPERATIONS BENCHMARK RESULTS
======================================================================

HOTPLUG:
  Success Rate:      10/10 (100.0%)
  Disks Attached:    30
  Disks Validated:   30
  Avg API Time:      1.25s
  Avg Volume Ready:  3.50s
  Avg Total Time:    7.20s

COLDPLUG:
  Success Rate:      10/10 (100.0%)
  Disks Attached:    30
  Disks Validated:   30
  Avg API Time:      0.80s
  Avg VM Boot Time:  45.00s
  Avg Total Time:    52.30s
======================================================================

View in Dashboard

cd dashboard
python3 generate_dashboard.py --base-dir ../results
open output/dashboard.html

The dashboard shows: - KPI cards (Hotplug/Coldplug avg time, success rate) - Time breakdown chart (API, Volume Ready/Boot, Validation) - Per-VM results table with validation status