Running a benchmark

Prerequisites

1. The device is visible under Devices
2. You're logged in as operator or admin (viewers cannot start runs)
3. No other run is in progress (Anvil serializes; one device at a time)

Start via UI

1. Devices → click your device → New run
2. Pick the profile
3. For destructive profiles, type the last 6 chars of the serial to confirm
4. Click Start

Start via API

bash TOKEN=$(grep ^ANVIL_BEARER_TOKEN .env | cut -d= -f2) curl -X POST -H "Authorization: Bearer $TOKEN" \ -H "Content-Type: application/json" \ -d '{ "device_id": "01K...", "profile_name": "snia_quick_pts", "confirm_serial_last6": "A1B2C3" }' \ http://localhost:8080/api/runs

Returns the new run's full record including its id. Poll GET /api/runs/{id} or subscribe to WS /ws/runs/{id} for live updates.

What Anvil captures per run

• Device fingerprint at run time (model, serial, firmware, capacity, PCIe capability, PCIe actual link)
• Host fingerprint (kernel, CPU, available fio / nvme versions)
• SMART before the first fio phase, SMART after the last
• Every fio json+ sample (per-second metrics per phase)
• Temperature from every SMART sample during the run

Aborting

Click Abort run on the Run Detail page. The queue sends SIGTERM to fio in the runner container; the run finishes its current sample, then writes status=aborted with error_message="aborted by user".

Thermal auto-abort

The endurance_soak profile watches the drive temperature every 5 seconds. If 6 consecutive samples report ≥ 75 °C, the run aborts itself with:

thermal_abort: temperature ≥ 75 °C for 6 consecutive SMART samples

This is deliberate — throttling kicks in on most NVMes near 75 °C and any data after that point is compromised as a benchmark result.