CentOS 7 BIOS/UEFI cloud images

[djuarezg]

CentOS 7 BIOS/UEFI cloud images

[szarkos]

Hi,

There is an attempt to create a CentOS 7.7 BIOS/UEFI kickstart at 1. I have not verified this config myself yet, I'll need to find time in the coming weeks. If you want to test it yourself then any feedback would be appreciated. Along with UEFI it looks like they are trying to make a transition to cloud-init as well in that kickstart, but aren't quite there yet. I would probably ignore that part.

Currently I have no plan to do the opposite - create a config to build the image in a BIOS boot environment. When we started down this path we followed closely what the Red Hat engineers recommended 2. In that bug report they tested both CentOS 7 and the CentOS 8 kickstart I provided and confirmed this process works well (after some additional tweaks). I'm not sure there is much appetite for testing again in a BIOS environment and dealing with any potential issues as we have a process now that's been well validated. Of course if you want to give it a shot it may be useful for some folks. Is there a reason you need to build in a BIOS env vs. UEFI?

[djuarezg]

@szarkos Basically we are using https://docs.pagure.org/koji/image_build/#building-disk-images which spawn BIOS VMs. Since there is no simple way for these VMs to spawn in UEFI mode, we wanted to replicate your kickstart in a BIOS environment to ease things.

Our final goal is to produce an Openstack CERN CentOS7 ready image (and 8 afterwards) that boots in both BIOS and UEFI VMs.

I'll take a look at the links you provided.

[djuarezg]

@szarkos Just one simple question. For this image building it asumes it will always run in UEFI VMs. Are you using Koji for such purpose? If so, do you have any reference on how you are forcing the UEFI boot on the spawned VMs?

[djuarezg]

OK, feel free to close this ticket once you publish the 7.7 ks. I could verify myself that it uses a working approach. I could be able to build it, but only after forcing UEFI on our spawned VMs for Koji.

Thank you for the references, they were really useful!

[djuarezg]

@szarkos Sorry to bother you again, but maybe you know why on the following lines:

• https://github.com/CentOS/sig-cloud-instance-build/blob/master/cloudimg/CentOS-8-x86_64-Azure.ks#L166
• https://github.com/openlogic/AzureBuildCentOS/blob/master/ks/azure/centos77.ks#L134

Even if dealing with BIOS Grub config (/boot/grub2/grub.cfg) it is pointing to a UEFI grubenv: (hd0,gpt15)/efi/centos/grubenv. Is this an attempt to keep a central grubenv for both BIOS and UEFI? This means that the EFI partition is always required on a BIOS node, am I right?

[szarkos]

Yes, that pointer to a single grubenv seems to be how the installer wants to set all this up anyway on a rhel/centos 8 system. In a previous iteration of this kickstart I tried copying the grubenv to both locations to maintain them separately, but then that introduced a bug where the kernel upgrade process did not update the default kernel for both uefi and bios grub configs properly. So the grubenv needs to be synced. And yes, the efi partition must always be there for this sort of bios/uefi install.

The folks from Red Hat helped me test and fixup this ks - there's more information in this public bug report starting at comment #26 and onward: https://bugzilla.redhat.com/show_bug.cgi?id=1570991#c26