get_licenses (napalm-device-types LicenseMixin, 4.15) on redfish and bmc,
for NetOrk/netork#1111.
- A BMC that refuses Redfish altogether reports each license it names as
missing, with its message.
- One that answers but withholds parts reports the license with withholds,
e.g. DCMS withholding drives and the firmware inventory on an X10 with
only SFT-OOB-LIC.
- ipmi knows no licenses.
get_device_warnings keeps its two codes until netOrk derives them.
bmc (napalm_redfish.bmc:BmcDriver) reads a BMC over Redfish where it
answers. A BMC that refuses Redfish (no license) is read and switched over
IPMI. A getter that fails over Redfish is asked over IPMI where IPMI has it;
IPMI is opened only then. Power control goes over Redfish when it answers,
and a refusal is final. The BMC is reset over Redfish only. Same account
for both; ipmi_port in optional_args.
Discovery suggests bmc for a BMC it finds: the fingerprint rules moved from
RedfishDriver to BmcDriver.
First run against real hardware over Redfish: an X10DRL-i (BMC 3.91) with
SFT-OOB-LIC. Its Redfish tree is recorded under
tests/fixtures/supermicro-x10drl-i-3.91/redfish, scrubbed.
- A resource refused for want of a further license (403 OemLicenseNotPassed)
no longer fails the getter. Storage then falls back to SimpleStorage. The
firmware list falls back to the BMC and BIOS versions when UpdateService
is withheld.
- get_device_warnings reports bmc_license_partial with the licenses and the
withheld parts (drives, firmware inventory).
- A placeholder system model ("Super Server") is replaced by the board
(PartNumber, else the chassis model).
IpmiDriver.set_power_state sends an IPMI chassis control: On, GracefulShutdown,
ForceOff and ForceRestart. get_power_state now offers what fits the server's
state: On when it is off, the other three when it is on. IPMI has no graceful
restart and pyghmi offers no power cycle, so neither is offered. The driver
still does not reset the BMC.
For boards that offer nothing but IPMI, such as a Supermicro X10 without a
Redfish license.
The second IPMI poll of a BMC in the same worker process failed with
"'<' not supported between instances of 'float' and 'NoneType'".
pyghmi files a session in Session.initting_sessions under the
credentials as str and deletes it only under bytes, so the entry stays
and the next Command got the old, logged-out session back. The driver
drops that entry itself before open() and after close(). Reproduced and
verified live against the Supermicro X10DRL-i: three polls in one
process now succeed.
IpmiDriver and RedfishDriver mix in HealthMetricsMixin, so netOrk's
SNMP health poll reads the BMC's own load, memory and interfaces where
the BMC runs net-snmp with the UCD-MIB, as the X10's does.
A Supermicro X10DRL-i refuses Redfish without a license (SUM, DCMS or
OOB), even the service root. IPMI answers without one.
- New entry point `ipmi` (napalm_redfish.ipmi.IpmiDriver) on
ManagementControllerDriver, over pyghmi: facts and inventory from the
FRU (serial number, system UUID), the BMC's LAN interface, sensors
(temperatures, fans, power supplies), the power state, the event log
and the BMC firmware. Read-only: no power control, no BMC reset.
Tested against recordings of the X10 with serials, the UUID, MACs and
IPs replaced.
- RedfishDriver.open() no longer fails on a license refusal: get_facts
returns what is known without Redfish, get_device_warnings reports
bmc_license_required with the licenses the BMC names, every other
getter raises. netOrk then shows the reason instead of an offline BMC.
- IpmiDriver is not exported from the package: NAPALM's own
get_network_driver("redfish") takes the first driver class of
napalm_redfish.
One Redfish answer (the service root or the error a service refuses it
with) and one landing page (Supermicro/ATEN, iDRAC or iLO) can match,
never all of them. Grouped, the driver's maximum in netOrk's discovery
is 21 instead of 41, and a Supermicro landing page alone (6 of 21)
outranks the "linux" its BMC reports over SNMP (5 of 18).
Requires napalm-device-types 4.8 for FingerprintRule.group.
netOrk's discovery took six switches for BMCs: the PortSpec for 443
added 5 points to every device with a web UI, and nothing else scored
them. The port now carries no weight.
The real BMC, a Supermicro X10DRL-i, was not recognised at all. Its web
UI (ATEN firmware) gets a rule, as do iDRAC and iLO; discovery fetches
those landing pages today. A rule for a Redfish error body covers the
service root once discovery reads it.
That board refuses Redfish without a Supermicro license, even the
service root (403 OemLicenseNotPassed, "SUM DCMS OOB"). The library
reports only the status, so open() reads the error body once more and
raises RedfishLicenseError naming the licenses.
Release 0.1.1.
The redfish driver implements get_facts, get_interfaces, get_interfaces_ip, get_environment, get_users, get_ntp_servers and get_snmp_information for the BMC, and the role methods get_power_state, set_power_state, get_hardware_inventory, get_event_log, get_firmware_inventory and get_management_protocols for the managed server. reboot_host resets the BMC through Manager.Reset, never the server.
Registered as the napalm.drivers entry point redfish; depends on napalm and napalm-device-types >= 4.6.
parse_power_supply also reads items of the legacy Power.PowerSupplies array, parse_ethernet_interface exposes the raw LinkStatus, and parse_network_protocol reports protocols that carry a Port without a ProtocolEnabled flag.
Pure functions for ServiceRoot, ComputerSystem (facts, power state, reset action and ActionInfo), Processors, Memory, Storage/Drives/SimpleStorage, Chassis Thermal and Power as well as ThermalSubsystem/PowerSubsystem/Sensors (with environment_sources to pick whichever exists), Manager, EthernetInterfaces, ManagerNetworkProtocol, accounts, firmware inventory and log services/entries.
Session login through the SessionService with X-Auth-Token, the session deleted on close (also on errors and when the DELETE fails), basic auth as fallback only when the service cannot create sessions, never after a rejected password. Certificate verification is on by default and configurable per device; True passes the requests CA bundle explicitly because the redfish library disables verification without a CA file.
Collections are followed by @odata.id, including expanded members and nextLink pages. Links to other hosts are refused so the session token stays on the BMC.
Unmodified files from DMTF Redfish-Publications (public-rackmount1, public-localstorage, public-telemetry), only those the tests use, each set with DMTF's BSD-3-Clause license and a COMMANDS.txt naming the source commit and request paths.
conftest.py loads them by request path and serves a mockup over mocked HTTP, with a SessionService that hands out a token, for the transport tests.
pyproject.toml (Apache-2.0, Python >= 3.10, depends on the DMTF redfish library) and an empty napalm_redfish package. No napalm.drivers entry point yet: the driver class waits for the ManagementControllerDriver role base.
No code yet. The README explains why one Redfish driver with vendor quirk
modules beats a Supermicro-only IPMI driver, and keeps IPMI over LAN as a
read-only fallback. docs/ holds a draft for a new ManagementController
role base in napalm-device-types, the getter-to-resource map, the first
test device (Supermicro X10DRL-i, Redfish firmware 3.91), the security
points (IPMI RAKP hash disclosure, power control, credentials) and the
test plan with DMTF mockups and fixtures to record.
Christian Manivong
christianmanivong