kayobe/doc/source/administration/bare-metal.rst

Baremetal Compute Node Management

When enrolling new hardware or performing maintenance, it can be useful to be able to manage many bare metal compute nodes simultaneously.

In all cases, commands are delegated to one of the controller hosts, and executed concurrently. Note that ansible's forks configuration option, which defaults to 5, may limit the number of nodes configured concurrently.

By default these commands wait for the state transition to complete for each node. This behavior can be changed by overriding the variable baremetal_compute_wait via -e baremetal_compute_wait=False

Register

This is an experimental workflow which acts as an alternative to enrolling nodes through inspection. Bare metal compute nodes can be registered in Ironic via Kayobe if defined in the Kayobe inventory. An example hosts file for group r1 is below:

[r1]
hv100 ipmi_address=1.2.3.4
...

[baremetal-compute:children]
r1

You should also define a group_vars file for this group containing the Ironic variables, this could be in etc/kayobe/inventory/group_vars/r1/ironic_vars or in the environment you are using.

ironic_driver: redfish

ironic_driver_info:
  redfish_system_id: "{{ ironic_redfish_system_id }}"
  redfish_address: "{{ ironic_redfish_address }}"
  redfish_username: "{{ ironic_redfish_username }}"
  redfish_password: "{{ ironic_redfish_password }}"
  redfish_verify_ca: "{{ ironic_redfish_verify_ca }}"
  ipmi_address: "{{ ipmi_address }}"

ironic_properties:
  capabilities: "{{ ironic_capabilities }}"

ironic_resource_class: "example_resource_class"
ironic_redfish_system_id: "/redfish/v1/Systems/System.Embedded.1"
ironic_redfish_verify_ca: "{{ inspector_rule_var_redfish_verify_ca }}"
ironic_redfish_address: "{{ ipmi_address }}"
ironic_redfish_username: "{{ inspector_redfish_username }}"
ironic_redfish_password: "{{ inspector_redfish_password }}"
ironic_capabilities: "boot_option:local,boot_mode:uefi"

It is essential that the Ironic username and password match the BMC username and password for your nodes. If the username and password combination is not the same for the entire group you will need to adjust your configuration accordingly. The IPMI address should also match the BMC address for your node.

Once this has been completed you can begin enrolling the Ironic nodes:

(kayobe) $ kayobe baremetal compute register

Inspector is not used to discover nodes and no node inspection will take place on enrollment, nodes will automatically be placed into manageable state. To inspect, you should use kayobe baremetal compute inspect following enrollment.

Manage

A node may need to be set to the manageable provision state in order to perform certain management operations, or when an enrolled node is transitioned into service. In order to manage a node, it must be in one of these states: enroll, available, cleaning, clean failed, adopt failed or inspect failed. To move the baremetal compute nodes to the manageable provision state:

(kayobe) $ kayobe baremetal compute manage

Provide

In order for nodes to be scheduled by nova, they must be available. To move the baremetal compute nodes from the manageable state to the available provision state:

(kayobe) $ kayobe baremetal compute provide

Inspect

Nodes must be in one of the following states: manageable, inspect failed, or available. To trigger hardware inspection on the baremetal compute nodes:

(kayobe) $ kayobe baremetal compute inspect

Saving Hardware Introspection Data

Introspection data will be stored in the nginx service within the inspection_store container. This data may be saved to the control host:

(kayobe) $ kayobe baremetal compute introspection data save

--output-dir may be used to specify the directory in which introspection data files will be saved. --output-format may be used to set the format of the files.

Rename

Once nodes have been discovered, it is helpful to associate them with a name to make them easier to work with. If you would like the nodes to be named according to their inventory host names, you can run the following command:

(kayobe) $ kayobe baremetal compute rename

This command will use the ipmi_address host variable from the inventory to map the inventory host name to the correct node.

Update Deployment Image

When the overcloud deployment images have been rebuilt or there has been a change to one of the following variables:

• ipa_kernel_upstream_url
• ipa_ramdisk_upstream_url

either by changing the url, or if the image to which they point has been changed, you need to update the deploy_ramdisk and deploy_kernel properties on the Ironic nodes. To do this you can run:

(kayobe) $ kayobe baremetal compute update deployment image

You can optionally limit the nodes in which this affects by setting baremetal-compute-limit:

(kayobe) $ kayobe baremetal compute update deployment image --baremetal-compute-limit sand-6-1

which should take the form of an ansible host pattern. This is matched against the Ironic node name.

Ironic Serial Console

To access the baremetal nodes from within Horizon you need to enable the serial console. For this to work the you must set kolla_enable_nova_serialconsole_proxy to true in etc/kayobe/kolla.yml:

kolla_enable_nova_serialconsole_proxy: true

The console interface on the Ironic nodes is expected to be ipmitool-socat, you can check this with:

openstack baremetal node show <node_id> --fields console_interface

where <node_id> should be the UUID or name of the Ironic node you want to check.

If you have set kolla_ironic_enabled_console_interfaces in etc/kayobe/ironic.yml, it should include ipmitool-socat in the list of enabled interfaces.

The playbook to enable the serial console currently only works if the Ironic node name matches the inventory hostname.

Once these requirements have been satisfied, you can run:

(kayobe) $ kayobe baremetal compute serial console enable

This will reserve a TCP port for each node to use for the serial console interface. The allocations are stored in ${KAYOBE_CONFIG_PATH}/console-allocation.yml. The current implementation uses a global pool, which is specified by ironic_serial_console_tcp_pool_start and ironic_serial_console_tcp_pool_end; these variables can set in etc/kayobe/ironic.yml.

To disable the serial console you can use:

(kayobe) $ kayobe baremetal compute serial console disable

The port allocated for each node is retained and must be manually removed from ${KAYOBE_CONFIG_PATH}/console-allocation.yml if you want it to be reused by another Ironic node with a different name.

You can optionally limit the nodes targeted by setting baremetal-compute-limit:

(kayobe) $ kayobe baremetal compute serial console enable --baremetal-compute-limit sand-6-1

which should take the form of an ansible host pattern.

Serial console auto-enable

To enable the serial consoles automatically on kayobe overcloud post configure, you can set ironic_serial_console_autoenable in etc/kayobe/ironic.yml:

ironic_serial_console_autoenable: true