Booting a Ramdisk or an ISO

Ironic supports booting a user provided ramdisk or an ISO image (starting with the Victoria release) instead of deploying a node. Most commonly this is performed when an instance is booted via PXE, iPXE or Virtual Media, with the only local storage contents being those in memory. It is suported by pxe, ipxe, redfish-virtual-media and ilo-virtual-media boot interfaces.


Ramdisk/ISO boot requires using the ramdisk deploy interface. As with most non-default interfaces, it must be enabled and set for a node to be utilized:

enabled_deploy_interfaces = iscsi,direct,ramdisk

Once enabled and the conductor(s) have been restarted, the interface can be set upon creation of a new node:

baremetal node create --driver ipmi \
    --deploy-interface ramdisk \
    --boot-interface ipxe

or update an existing node:

baremetal node set <NODE> --deploy-interface ramdisk

You can also use it with redfish virtual media instead of iPXE.

Creating a ramdisk

A ramdisk can be created using the ironic-ramdisk-base element from ironic-python-agent-builder, e.g. with Debian:

export ELEMENTS_PATH=/opt/stack/ironic-python-agent-builder/dib
disk-image-create -o /output/ramdisk \
    debian-minimal ironic-ramdisk-base openssh-server dhcp-all-interfaces

You should consider using the following elements:

The resulting files (/output/ramdisk.kernel and /output/ramdisk.initramfs in this case) can then be used when Booting a ramdisk.

Booting a ramdisk

Pass the kernel and ramdisk as normally, also providing the ramdisk as an image source, for example,

baremetal node set <NODE> \
    --instance-info kernel=http://path/to/ramdisk.kernel \
    --instance-info ramdisk=http://path/to/ramdisk.initramfs \
    --instance-info image_source=http://path/to/ramdisk.initramfs
baremetal node deploy <NODE>


The requirement to pass image_source is artificial and will be fixed in a future version of the Bare Metal service.

Booting an ISO

The ramdisk deploy interface can also be used to boot an ISO image. For example,

baremetal node set <NODE> \
    --instance-info boot_iso=http://path/to/boot.iso
baremetal node deploy <NODE>


This feature, when utilized with the ipxe boot_interface, will only allow a kernel and ramdisk to be booted from the supplied ISO file. Any additional contents, such as additional ramdisk contents or installer package files will be unavailable after the boot of the Operating System. Operators wishing to leverage this functionality for actions such as OS installation should explore use of the standard ramdisk deploy_interface along with the instance_info/kernel_append_params setting to pass arbitrary settings such as a mirror URL for the initial ramdisk to load data from. This is a limitation of iPXE and the overall boot process of the operating system where memory allocated by iPXE is released.


The intended use case is for advanced scientific and ephemeral workloads where the step of writing an image to the local storage is not required or desired. As such, this interface does come with several caveats:

  • Configuration drives are not supported with network boot, only with Redfish virtual media.

  • Disk image contents are not written to the bare metal node.

  • Users and Operators who intend to leverage this interface should expect to leverage a metadata service, custom ramdisk images, or the instance_info/ramdisk_kernel_arguments parameter to add options to the kernel boot command line.

  • When using PXE/iPXE boot, bare metal nodes must continue to have network access to PXE and iPXE network resources. This is contrary to most tenant networking enabled configurations where this access is restricted to the provisioning and cleaning networks

  • As with all deployment interfaces, automatic cleaning of the node will still occur with the contents of any local storage being wiped between deployments.