Regression: BIOS boot detection broken, when booting via iPXE

[trickkiste]

Bug Report

Regression from v1.11.5 -> v1.12.1: Booting a kernel initrd Talos installer via iPXE, results in a system which does have a BIOS partition but does not boot from it.

Description

I have an old HP ProLiant DL320e Gen8 v2 which does not come with UEFI nor secureboot. I was able to install Talos v1.11.5 with no problem, but doing the same on v1.12.1 results in an unbootable system. It seems the right partitions are created but GRUB is not being installed.

The system does not boot from disk and therefore reverts back to booting via iPXE into the installer again.

After the server boots from iPXE again, I can see there are EFI BIOS META BOOT partitions now. But I see no STATE and no EPHEMERAL partitions. I guess they are created at first boot of the installed system, which unfortunately does not happen.

It would be nice, if I could FORCE a GRUB install, but there are no config options for that.

If I remember correctly iPXE emulates an UEFI boot. And it seems Talos relies on that in the end, to decide against grub-install but uses some other way do determine if this is a BIOS system to decide the partition layout, since the right partitions are there but GRUB is not!

It appears the CD-ROM version has the same problem, at least when booted via iLO4 virtual media functionality.

Environment

• Talos version: v1.12.1
• Platform: metal

[smira]

There were no changes around GRUB/UEFI, you need to capture and look into the install logs.

If you see BIOS + BOOT partitions, Talos was installed with GRUB support.

iPXE mode depends on how you boot it, it has nothing to do with Talos (UEFI/BIOS).

[trickkiste]

@smira There must be a difference between versions 1.11.6 and 1.12.1. I just setup my cluster with version 1.11.6 again and the BIOS nodes in question install and boot just fine.

Maybe it's something that appears insignificant but has an unexpected impact somewhere else. I have tried to get the nodes running with 1.12.1 a dozen times yesterday. Either setup does not finish as a whole or just the bootloader is not applied in the end.

But this problem only presents itself on BIOS nodes but not on my UEFI nodes. Which parts of Talos code might play a role in this?

[smira]

We can't say without the logs. If you provide the logs, we can look into this. Otherwise not much we can do.