Redfish driver

Overview

The redfish driver enables managing servers compliant with the Redfish protocol.

Prerequisites

  • The Sushy library should be installed on the ironic conductor node(s).

    For example, it can be installed with pip:

    sudo pip install sushy
    

Enabling the Redfish driver

  1. Add redfish to the list of enabled_hardware_types, enabled_power_interfaces, enabled_management_interfaces and enabled_inspect_interfaces as well as redfish-virtual-media to enabled_boot_interfaces in /etc/ironic/ironic.conf. For example:

    [DEFAULT]
    ...
    enabled_hardware_types = ipmi,redfish
    enabled_boot_interfaces = ipxe,redfish-virtual-media
    enabled_power_interfaces = ipmitool,redfish
    enabled_management_interfaces = ipmitool,redfish
    enabled_inspect_interfaces = inspector,redfish
    
  2. Restart the ironic conductor service:

    sudo service ironic-conductor restart
    
    # Or, for RDO:
    sudo systemctl restart openstack-ironic-conductor
    

Registering a node with the Redfish driver

Nodes configured to use the driver should have the driver property set to redfish.

The following properties are specified in the node’s driver_info field:

  • redfish_address: The URL address to the Redfish controller. It must

    include the authority portion of the URL, and can optionally include the scheme. If the scheme is missing, https is assumed. For example: https://mgmt.vendor.com. This is required.

  • redfish_system_id: The canonical path to the ComputerSystem resource

    that the driver will interact with. It should include the root service, version and the unique resource path to the ComputerSystem. This property is only required if target BMC manages more than one ComputerSystem. Otherwise ironic will pick the only available ComputerSystem automatically. For example: /redfish/v1/Systems/1.

  • redfish_username: User account with admin/server-profile access

    privilege. Although not required, it is highly recommended.

  • redfish_password: User account password. Although not required, it is

    highly recommended.

  • redfish_verify_ca: If redfish_address has the https scheme, the

    driver will use a secure (TLS) connection when talking to the Redfish controller. By default (if this is not set or set to True), the driver will try to verify the host certificates. This can be set to the path of a certificate file or directory with trusted certificates that the driver will use for verification. To disable verifying TLS, set this to False. This is optional.

  • redfish_auth_type: Redfish HTTP client authentication method. Can be

    “basic”, “session” or “auto”. The “auto” mode first tries “session” and falls back to “basic” if session authentication is not supported by the Redfish BMC. Default is set in ironic config as [redfish]auth_type.

The baremetal node create command can be used to enroll a node with the redfish driver. For example:

baremetal node create --driver redfish --driver-info \
  redfish_address=https://example.com --driver-info \
  redfish_system_id=/redfish/v1/Systems/CX34R87 --driver-info \
  redfish_username=admin --driver-info redfish_password=password \
  --name node-0

For more information about enrolling nodes see Enrollment in the install guide.

Boot mode support

The redfish hardware type can read current boot mode from the bare metal node as well as set it to either Legacy BIOS or UEFI.

Note

Boot mode management is the optional part of the Redfish specification. Not all Redfish-compliant BMCs might implement it. In that case it remains the responsibility of the operator to configure proper boot mode to their bare metal nodes.

UEFI secure boot

Secure boot mode can be automatically set and unset during deployment for nodes in UEFI boot mode, see UEFI secure boot mode for an explanation how to use it.

Two clean and deploy steps are provided for key management:

management.reset_secure_boot_keys_to_default

resets secure boot keys to their manufacturing defaults.

management.clear_secure_boot_keys

removes all secure boot keys from the node.

Out-Of-Band inspection

The redfish hardware type can inspect the bare metal node by querying Redfish compatible BMC. This process is quick and reliable compared to the way the inspector hardware type works i.e. booting bare metal node into the introspection ramdisk.

Note

The redfish inspect interface relies on the optional parts of the Redfish specification. Not all Redfish-compliant BMCs might serve the required information, in which case bare metal node inspection will fail.

Note

The local_gb property cannot always be discovered, for example, when a node does not have local storage or the Redfish implementation does not support the required schema. In this case the property will be set to 0.

Virtual media boot

The idea behind virtual media boot is that BMC gets hold of the boot image one way or the other (e.g. by HTTP GET, other methods are defined in the standard), then “inserts” it into node’s virtual drive as if it was burnt on a physical CD/DVD. The node can then boot from that virtual drive into the operating system residing on the image.

The major advantage of virtual media boot feature is that potentially unreliable TFTP image transfer phase of PXE protocol suite is fully eliminated.

Hardware types based on the redfish fully support booting deploy/rescue and user images over virtual media. Ironic builds bootable ISO images, for either UEFI or BIOS (Legacy) boot modes, at the moment of node deployment out of kernel and ramdisk images associated with the ironic node.

To boot a node managed by redfish hardware type over virtual media using BIOS boot mode, it suffice to set ironic boot interface to redfish-virtual-media, as opposed to ipmitool.

baremetal node set --boot-interface redfish-virtual-media node-0

Warning

Dell hardware requires a non-standard Redfish call to boot from virtual media, thus you must use the idrac hardware type and the idrac-redfish-virtual-media boot interface with it instead. See iDRAC driver for more details on this hardware type.

If UEFI boot mode is desired, the user should additionally supply EFI System Partition image (ESP), see Configuring an ESP image for details.

If [driver_info]/config_via_floppy boolean property of the node is set to true, ironic will create a file with runtime configuration parameters, place into on a FAT image, then insert the image into node’s virtual floppy drive.

When booting over PXE or virtual media, and user instance requires some specific kernel configuration, [instance_info]/kernel_append_params property can be used to pass user-specified kernel command line parameters. For ramdisk kernel, [instance_info]/kernel_append_params property serves the same purpose.

Pre-built ISO images

By default an ISO images is built per node using the deploy kernel and initramfs provided in the configuration or the node’s driver_info. Starting with the Wallaby release it’s possible to provide a pre-built ISO image:

baremetal node set node-0 \
  --driver_info redfish_deploy_iso=http://url/of/deploy.iso \
  --driver_info redfish_rescue_iso=http://url/of/rescue.iso

Note

OpenStack Image service (glance) image IDs and file:// links are also accepted.

No customization is currently done to the image, so e.g. Layer 3 or DHCP-less ramdisk booting won’t work. Configuring an ESP image is also unnecessary.

Configuring an ESP image

An ESP image is an image that contains the necessary bootloader to boot the ISO in UEFI mode. You will need a GRUB2 image file, as well as Shim for secure boot. See UEFI PXE - Grub setup for an explanation how to get them.

Then the following script can be used to build an ESP image:

DEST=/path/to/esp.img
GRUB2=/path/to/grub.efi
SHIM=/path/to/shim.efi
TEMP_MOUNT=$(mktemp -d)

dd if=/dev/zero of=$DEST bs=4096 count=1024
mkfs.fat -s 4 -r 512 -S 4096 $DEST

sudo mount $DEST $TEMP_MOUNT
sudo mkdir -p $DEST/EFI/BOOT
sudo cp "$SHIM" $DEST/EFI/BOOT/BOOTX64.efi
sudo cp "$GRUB2" $DEST/EFI/BOOT/GRUBX64.efi
sudo umount $TEMP_MOUNT

Note

If you use an architecture other than x86-64, you’ll need to adjust the destination paths.

The resulting image should be provided via the driver_info/bootloader ironic node property in form of an image UUID or a URL:

baremetal node set --driver-info bootloader=<glance-uuid-or-url> node-0

Alternatively, set the bootloader UUID or URL in the configuration file:

[conductor]
bootloader = <glance-uuid-or-url>

Finally, you need to provide the correct GRUB2 configuration path for your image. In most cases this path will depend on your distribution, more precisely, the distribution you took the GRUB2 image from. For example:

CentOS:

[DEFAULT]
grub_config_path = EFI/centos/grub.cfg

Ubuntu:

[DEFAULT]
grub_config_path = EFI/ubuntu/grub.cfg

Note

Unlike in the script above, these paths are case-sensitive!

Virtual Media Ramdisk

The ramdisk deploy interface can be used in concert with the redfish-virtual-media boot interface to facilitate the boot of a remote node utilizing pre-supplied virtual media. See Booting a Ramdisk or an ISO for information on how to enable and configure it.

Instead of supplying an [instance_info]/image_source parameter, a [instance_info]/boot_iso parameter can be supplied. The image will be downloaded by the conductor, and the instance will be booted using the supplied ISO image. In accordance with the ramdisk deployment interface behavior, once booted the machine will have a provision_state of ACTIVE.

baremetal node set <node name or UUID> \
    --boot-interface redfish-virtual-media \
    --deploy-interface ramdisk \
    --instance_info boot_iso=http://url/to.iso

This initial interface does not support bootloader configuration parameter injection, as such the [instance_info]/kernel_append_params setting is ignored.

Configuration drives are supported starting with the Wallaby release for nodes that have a free virtual USB slot:

baremetal node deploy <node name or UUID> \
    --config-drive '{"meta_data": {...}, "user_data": "..."}'

or via a link to a raw image:

baremetal node deploy <node name or UUID> \
    --config-drive http://example.com/config.img

Layer 3 or DHCP-less ramdisk booting

DHCP-less deploy is supported by the Redfish virtual media boot. See Layer 3 or DHCP-less ramdisk booting for more information.

Firmware update using manual cleaning

The redfish hardware type supports updating the firmware on nodes using a manual cleaning step.

The firmware update cleaning step allows one or more firmware updates to be applied to a node. If multiple updates are specified, then they are applied sequentially in the order given. The server is rebooted once per update. If a failure occurs, the cleaning step immediately fails which may result in some updates not being applied. If the node is placed into maintenance mode while a firmware update cleaning step is running that is performing multiple firmware updates, the update in progress will complete, and processing of the remaining updates will pause. When the node is taken out of maintenance mode, processing of the remaining updates will continue.

When updating the BMC firmware, the BMC may become unavailable for a period of time as it resets. In this case, it may be desireable to have the cleaning step wait after the update has been applied before indicating that the update was successful. This allows the BMC time to fully reset before further operations are carried out against it. To cause the cleaning step to wait after applying an update, an optional wait argument may be specified in the firmware image dictionary. The value of this argument indicates the number of seconds to wait following the update. If the wait argument is not specified, then this is equivalent to wait 0, meaning that it will not wait and immediately proceed with the next firmware update if there is one, or complete the cleaning step if not.

The update_firmware cleaning step accepts JSON in the following format:

[{
    "interface": "management",
    "step": "update_firmware",
    "args": {
        "firmware_images":[
            {
                "url": "<url_to_firmware_image1>",
                "wait": <number_of_seconds_to_wait>
            },
            {
                "url": "<url_to_firmware_image2>"
            },
            ...
        ]
    }
}]

The different attributes of the update_firmware cleaning step are as follows:

Attribute

Description

interface

Interface of the cleaning step. Must be management for firmware update

step

Name of cleaning step. Must be update_firmware for firmware update

args

Keyword-argument entry (<name>: <value>) being passed to cleaning step

args.firmware_images

Ordered list of dictionaries of firmware images to be applied

Each firmware image dictionary, is of the form:

{
  "url": "<URL of firmware image file>",
  "wait": <Optional time in seconds to wait after applying update>
}

The url argument in the firmware image dictionary is mandatory, while the wait argument is optional.

Note

Only http and https URLs are currently supported in the url argument.

Note

At the present time, targets for the firmware update cannot be specified. In testing, the BMC applied the update to all applicable targets on the node. It is assumed that the BMC knows what components a given firmware image is applicable to.

To perform a firmware update, first download the firmware to a web server that the BMC has network access to. This could be the ironic conductor web server or another web server on the BMC network. Using a web browser, curl, or similar tool on a server that has network access to the BMC, try downloading the firmware to verify that the URLs are correct and that the web server is configured properly.

Next, construct the JSON for the firmware update cleaning step to be executed. When launching the firmware update, the JSON may be specified on the command line directly or in a file. The following example shows one cleaning step that installs two firmware updates. The first updates the BMC firmware followed by a five minute wait to allow the BMC time to start back up. The second updates the firmware on all applicable NICs.:

[{
    "interface": "management",
    "step": "update_firmware",
    "args": {
        "firmware_images":[
            {
                "url": "http://192.0.2.10/BMC_4_22_00_00.EXE",
                "wait": 300
            },
            {
                "url": "https://192.0.2.10/NIC_19.0.12_A00.EXE"
            }
        ]
    }
}]

Finally, launch the firmware update cleaning step against the node. The following example assumes the above JSON is in a file named firmware_update.json:

baremetal node clean <ironic_node_uuid> --clean-steps firmware_update.json

In the following example, the JSON is specified directly on the command line:

baremetal node clean <ironic_node_uuid> --clean-steps '[{"interface": "management", "step": "update_firmware", "args": {"firmware_images":[{"url": "http://192.0.2.10/BMC_4_22_00_00.EXE", "wait": 300}, {"url": "https://192.0.2.10/NIC_19.0.12_A00.EXE"}]}}]'

Note

Firmware updates may take some time to complete. If a firmware update cleaning step consistently times out, then consider performing fewer firmware updates in the cleaning step or increasing clean_callback_timeout in ironic.conf to increase the timeout value.

Warning

Warning: Removing power from a server while it is in the process of updating firmware may result in devices in the server, or the server itself becoming inoperable.