Packer Created VMs Failing to Boot

dan89

When creating a Windows VM with Packer using UEFI boot firmware, the VM is failing to boot around 88% with no errors logged in XOA.

I am using the following Packer plugin:
https://github.com/ddelnano/packer-plugin-xenserver/tree/main

Things tested/checked so far:

If the VM is switched from UEFI to BIOS it boots but then fails to boot if switched back to UEFI.
If a VM is manually created in XOA from the same template it boots with no issue
If the firmware="uefi" option is removed from Packer, the VM defaults to BIOS and boots fine. However, if the VM is switched to UEFI it fails to boot
Using Packer to create a UEFI CentOS VM works as expected
Re-created the cloned Windows template and issue persists

I checked the output of xe vm-list with all params, and the only major differences I noticed are:

Packer created VM:

platform (MRW): timeoffset: 0; device-model: qemu-upstream-uefi; viridian: true

other-config (MRW): xo:03af740a: {"creation":{"date":"2023-11-06T20:16:30.631Z","template":"1c33af1c-e919-418c-ad45-85d7d6fb604a","user":"f8241546-6b58-484e-92d0-58817ea45b3f"}}; base_template_name: Windows Server 2022 (64-bit); import_task: OpaqueRef:55b37ba3-84a2-47ac-b8a6-41fb63a69ebc; mac_seed: ceaf53ff-f229-63d2-b586-5cb5aaea6686; install-methods: cdrom

XOA created VM:

platform (MRW): secureboot: false; device-model: qemu-upstream-uefi; timeoffset: 0; videoram: 8; hpet: true; viridian_apic_assist: true; apic: true; device_id: 0002; cores-per-socket: 2; viridian_crash_ctl: true; pae: true; vga: std; nx: true; viridian_time_ref_count: true; viridian_stimer: true; viridian: true; acpi: 1; viridian_reference_tsc: true

other-config (MRW): xo:17504efc: {"creation":{"date":"2023-11-06T21:03:58.450Z","template":"03af740a-ca0d-4246-e64d-bab99b90f682","user":"f8241546-6b58-484e-92d0-58817ea45b3f"}}; xo:03af740a: {"creation":{"date":"2023-11-06T20:16:30.631Z","template":"1c33af1c-e919-418c-ad45-85d7d6fb604a","user":"f8241546-6b58-484e-92d0-58817ea45b3f"}}; base_template_name: Windows Server 2022 (64-bit); import_task: OpaqueRef:55b37ba3-84a2-47ac-b8a6-41fb63a69ebc; mac_seed: f5d234ad-6d5c-012b-a2f7-2c2ee9a61cca; install-methods: cdrom

olivierlambert

You are probably not using the right template in the first place

dan89

@olivierlambert That is always a possibility

I created the Windows and CentOS templates the same way by creating a base VM without installing the OS and then capturing that as a Template. I also tried copying the default pre-defined Windows Server 2022 template.

olivierlambert

@AtaxyaNetwork or @ddelnano might tell you how to select the right template

dan89

@olivierlambert - Figured it out for anyone else's future reference.

When setting platform_args you need to define all of them even if you only need to change one.

Since the defaults are

{
    "viridian": "false",
    "nx": "true",
    "pae": "true",
    "apic": "true",
    "timeoffset": "0",
    "acpi": "1",
    "cores-per-socket": "1"
}

I needed to include all of them even though I only needed "viridian": "true". Once I added in the other options the VM booted and the build worked.

AtaxyaNetwork

Hi @dan89 !

I'm not a windows expert, but I can do some testing.

Can you show us your pkr.hcl file ?

AtaxyaNetwork

We posted at the same time

@dan89 Glad you figured it out !

Maybe it can be documented in the packer repo. What do you think @ddelnano ?

mtcoffee

@dan89

Do you have a pkr.hcl for windows that you can share?

I am curious how your handling winrm and autounattend.xml?

fatek

@mtcoffee said in Packer Created VMs Failing to Boot:

Do you have a pkr.hcl for windows that you can share?

Follow up question: Do you have a pkr.hcl for windows that you can share?

mtcoffee

@fatek

After posting this I ended up building out a sample project and shared on github.
https://github.com/mtcoffee/xcp-ng-packer-examples

fatek

@mtcoffee talk about great timing?
About 25 minutes ago, I git cloned your repo!
Thanks.

fatek

@mtcoffee Any luck getting autounattend.xml - UEFI to work?

mtcoffee

@fatek

I noted that in the packer file. I was not able to get packers floppy_files working in uefi mode.

@ddelnano might have some insight?

fatek

Anyone know where the @ddelnano Discord channel is?

olivierlambert

https://discord.gg/Cp2CDhqd

fatek

@olivierlambert thank you

fatek

I would really love to get this working w XCP, no bios, I mean uefi only.
Packer was super easy to setup on ESXi + vCenter!

olivierlambert

Contributions are welcome, we are pretty busy here, and if we planned to have a fully dedicated "Devops" team, it will be in place in September. So please be patient, we do not have 5000 engineers

mtcoffee

@olivierlambert is spot on. Manage expectations. The best thing we can do is contribute helpful information.

Here is what I was able to deduce when this issue came up.

When creating a Windows VM in uefi mode and using the floppy_file option in packer:

There is a disk added as "fda"
However the fda device is not available as the a: drive or any other drive letter. When booted using bios mode, you can find the fda device by settting drive letter to "a:".

This is likely because uefi typically does not support floppy disks. I believe in VMWare when the floppy_file option was used in UEFI mode, it added the ISO as a DVD drive instead of an FDA.

Not sure if this is best solved at the Packer level or if it should be intercepted and corrected XCP-ng level?

fatek

I hear you regarding "managing expectations".

If Vates (as a company) is going to claim they offer VMware type/like value then Windows can not be a 2nd class citizen in your ecosystem!