Extend OS Installation Options for Machines

[hardikdr]

Summary

Add support for multiple boot modes concerning how operating-system is installed on the Machine.

• Installation of the OS directly on the disk.
• Installation of the OS in memory for the session.
• Preserving specific bind mounts post-boot.

Progress Tracker:

• Develop the initial version of the on-disk installation mechanism. @5kt
• Enhance the installer by integrating it with the initrd. @5kt
• Publish updated operating system images that include the new initrd. @hardikdr
• Design and implement APIs for incorporating the installer into the Metal Operator.

Motivation

Better flexibility in deployment scenarios.

[hardikdr]

A possible enhancement in the APIs could look following, it's open for discussion:

ServerClaim:

apiVersion: metal.ironcore.dev/v1alpha1
kind: ServerClaim
metadata:
  name: example-serverclaim
spec:
  power: "On"
  ...
  bootPolicy: NetworkBootOnce   # Options: NetworkBootOnce / NetworkBootAlways / None
status:
  conditions:
    - type: NetworkBootedOnce
      status: "True"
      reason: Network boot was performed once.

• bootPolicy Field in ServerClaim: Add a spec.bootPolicy field with the following values:

  • NetworkBootOnce: Set PXEBootOnce signal for the first boot, and delete the associated ServerBootConfig after the successful power cycle operation execution.
  • NetworkBootAlways: Always set a PXEBootOnce signal during power cycles and retain the ServerBootConfig.
  • None: Do not set any boot-related signals, and do not create a ServerBootConfig.

• ServerBootConfig Lifecycle:

  • Delete ServerBootConfig after the first boot for NetworkBootOnce.
  • Retain ServerBootConfig indefinitely for NetworkBootAlways.
  • Skip creation of ServerBootConfig for None.

• Tracking Boot State:

  • Add a NetworkBootedOnce condition in ServerClaim.status to track if the network boot was successfully performed.

cc @5kt @defo89 @stefanhipfel @afritzler

[defo89]

Thanks @hardikdr! I like the proposal. With above implemented, how would you approach behaviour in #175 ?

I guess for Reserved server with bootPolicy: NetworkBootOnce it would just reconcile fully (with PXE boot) once and since ServerBootConfig is deleted afterwards it would not go through this in further reconciliations?