A fast, copy-paste runbook to validate the Redfish driver — especially the new boot-device control (BootSourceOverride, #28/#201) and power — against two real BMCs. Tracks epic #200; capture findings on #29 (real-BMC validation) and the support matrix (#96). Pairs with the synthetic coverage in tests/test_redfish_boot.py / tests/redfish_emulator.py.

Status: the R710/iDRAC6 leg was executed 2026-07-14 — the IPMI driver (#62, shipped in v0.1.0b7) validated live end-to-end (power, boot-device, sensors, SEL, SOL console; results on #206 and the Hardware-Compatibility list). The DL380 G9 / iLO4 Redfish leg (#29) is still open — the runbook below remains the procedure for it, and a reusable template for similar boxes.

Read the firmware-era caveat first — it decides which box is testable over Redfish.

So: full Redfish run on the DL380 G9; on the R710, confirm clean detection of "no Redfish", then run the same phases through the IPMI driver (see the R710 section below). If the R710's iDRAC was flashed to something newer, or it's actually a 12th-gen board, re-check — but plan for iDRAC6 = no Redfish.

• kvm-pilot installed (uv run kvm-pilot ... in the repo, or pip install).
• Network reachability from the runner to each BMC IP (BMCs are on a dedicated mgmt NIC/VLAN — confirm the runner can reach it; the same split-tunnel/L2 gotcha that hid 10.0.1.16 applies).
• Credentials. Do not commit them. Defaults to try: iLO4 Administrator / the label password on the server pull-tab; iDRAC root / calvin.
• A maintenance window — several steps power-cycle the host.

Add to ~/.config/kvm-pilot/config.toml (password via KVM_PILOT_PASSWD, --passwd-file <600 file>, or --ask-passwd — never in the file):

[hosts.ilo-dl380g9]
host = "10.0.1.AA"        # iLO IP
driver = "redfish"
user = "Administrator"
verify_ssl = false         # iLO4 ships a self-signed cert; pin with ssl_ca_file if you have it

[hosts.idrac-r710]
host = "10.0.1.BB"        # iDRAC IP
driver = "ipmi"            # iDRAC6 has no Redfish; shells out to ipmitool
user = "root"

Then per session: export KVM_PILOT_PASSWD='...' (or pass --ask-passwd).

P=ilo-dl380g9   # then repeat with P=idrac-r710
uv run kvm-pilot healthcheck   --profile $P     # intake gate — expect no CRITICAL for a reachable BMC
uv run kvm-pilot info          --profile $P     # manufacturer/model/BIOS/power_state/redfish_version
uv run kvm-pilot capabilities  --profile $P     # expect: system_info, power, boot_progress, sensors, logs, virtual_media, boot_config
uv run kvm-pilot power-state   --profile $P
uv run kvm-pilot boot-device   --profile $P --show   # current override + ALLOWABLE targets + mode_settable
uv run kvm-pilot sensors       --profile $P     # temps/fans/power (iLO4/iDRAC expose these)
uv run kvm-pilot boot-progress --profile $P
uv run kvm-pilot logs          --profile $P --seek 3600

Record from info / boot-device --show: redfish_version, the allowable boot targets, and mode_settable (does the box expose BootSourceOverrideMode?). These drive the quirks table below.

--show is read-only; every set needs --yes (gated) and writes BootSourceOverride. Verify each with a follow-up --show.

P=ilo-dl380g9
# one-time PXE (default: once + UEFI)
uv run kvm-pilot boot-device pxe  --profile $P --yes
uv run kvm-pilot boot-device      --profile $P --show      # expect enabled=Once, target=pxe

# one-time CD/virtual-media, then HDD
uv run kvm-pilot boot-device cd   --profile $P --yes
uv run kvm-pilot boot-device hdd  --profile $P --yes

# persistent + legacy variants (exercise both flags)
uv run kvm-pilot boot-device pxe  --profile $P --persistent --yes   # enabled=Continuous
uv run kvm-pilot boot-device bios --profile $P --legacy --yes       # BootSourceOverrideMode=Legacy (if settable)

# clear the override
uv run kvm-pilot boot-device none --profile $P --yes                # enabled=Disabled

End-to-end confirmation (optional, needs a reboot): set pxe once, then power reset --profile $P --yes, and confirm on the console/iLO that it PXE-boots once and reverts afterward.

Watch for (and note on #29):

• A target in your list that the BMC rejects → the driver should raise a clear CapabilityError naming the allowable set (not an opaque 400).
• iLO4 rejecting BootSourceOverrideMode → the driver auto-retries without it (log line: "rejected BootSourceOverrideMode; retrying without it"); confirm the target still applied.
• A 202 Accepted + Task on the PATCH → the driver polls it to completion.

P=ilo-dl380g9
uv run kvm-pilot power-state  --profile $P
uv run kvm-pilot power on     --profile $P --yes    # verified against Redfish PowerState
uv run kvm-pilot power off    --profile $P --yes    # graceful (GracefulShutdown)
uv run kvm-pilot power reset  --profile $P --yes
uv run kvm-pilot power-cycle  --profile $P --yes    # off-hard -> on, blocks on PowerState

Redfish power is verified (real PowerState), unlike GL ATX. Note the ResetTypes the box advertises (iLO4 vs iDRAC differ) and any InvalidOperationForSystemState 400/409 the driver absorbs as success.

uv run kvm-pilot media-list --profile ilo-dl380g9
# uv run kvm-pilot mount <http-url-to.iso> --profile ilo-dl380g9 --yes
# uv run kvm-pilot eject --profile ilo-dl380g9 --yes

iLO4 needs an advanced/iLO license for scriptable virtual media — note if it's absent. Combine with boot-device cd for a full remote-install rehearsal.

For the host's OS NIC (not the BMC): confirm ethtool <if> shows Wake-on: g, note the wired MAC, suspend/shutdown the OS, then from a sender on the same L2:

uv run python -c "from kvm_pilot import wol; wol.send_magic_packet('AA:BB:CC:DD:EE:FF', broadcast='10.0.1.255')"

(Already hardware-validated on the RHEL host behind KVM .20 — woke in 52 s.)

iDRAC6 has no Redfish (a probe at /redfish/v1 fails with 404 / connection reset) — the box runs through the IPMI driver (#62, driver = "ipmi", shells out to the system ipmitool). The same phase structure applies:

P=idrac-r710
uv run kvm-pilot healthcheck   --profile $P     # intake gate
uv run kvm-pilot info          --profile $P     # model from FRU Board Product (#206)
uv run kvm-pilot power-state   --profile $P
uv run kvm-pilot sensors       --profile $P     # temps/fans/PSU via SDR
uv run kvm-pilot logs          --profile $P     # SEL
uv run kvm-pilot boot-device pxe --profile $P --yes   # then cd/hdd/none, as Phase B
uv run kvm-pilot power reset   --profile $P --yes     # as Phase C
uv run kvm-pilot console       --profile $P            # SOL; exit with ~. (#208)

All of the above were live-validated on the R710 (14 pass / 0 fail) on 2026-07-14, including a SOL boot capture and an F11 boot-menu drive. Quirks found: iDRAC6 maps SOL to COM2 (ttyS1 on the host), and the console is text-only — no graphical/GUI installer phases over SOL.

For each box, record:

• RedfishVersion and BMC firmware version
• capabilities() set actually served
• boot-device allowable targets + whether mode_settable
• Did BootSourceOverrideMode apply, get rejected (auto-retry), or absent?
• one-time vs persistent both honored? cleared cleanly?
• PATCH sync (200/204) vs async (202+Task)?
• Power ResetTypes advertised; any 400/409-absorbed-as-success
• Virtual media present/licensed?
• Auth: session vs basic; session-cap/idle-timeout behavior
• Any field/shape that differs from the emulator (feed it back into redfish_emulator.py)