[CI][C9][Master][UEFI] ipxe boot doesn't pick up bits for provisioning

[Dariusz Smigiel]

Running side-by-side legacy ipxe and uefi ipxe revealed problems with provisioning OVB nodes.

UEFI test project [1], running with "ipxe-uefi-boot" [2] with extra properties: hw_firmware_type='uefi', hw_machine_type='q35' failed provisioning [3] being stuck on "UEFI boot" (right side of the image) [4].

For a comparison, legacy (non-UEFI) boot [5] went through with provisioning. It failed later on unrelated issue.

[1]: https://review.rdoproject.org/r/c/testproject/+/40332
[2]: https://paste.opendev.org/show/bBcJ7x1pBvKLwUSFRTwc/
[3]: https://logserver.rdoproject.org/32/40332/2/check/periodic-tripleo-ci-centos-9-ovb-3ctlr_1comp-featureset001-master/7068779/
[4]: https://i.imgur.com/of0Xi5F.png
[5]: https://review.rdoproject.org/r/c/testproject/+/40366

[Dariusz Smigiel (smigiel-dariusz)]

I tested on periodic-tripleo-ci-centos-8-ovb-3ctlr_1comp-featureset001-wallaby [1] and the issue seems to be the same.
Overcloud doesn't work through an introspection.

[1]: https://logserver.rdoproject.org/32/40332/4/check/periodic-tripleo-ci-centos-8-ovb-3ctlr_1comp-featureset001-wallaby/b0f128e/logs/undercloud/home/zuul/overcloud_introspect.log.txt.gz

[Cédric Jeanneret (cjeanner)]

Quick notes:
- latest code is apparently working fine on a fully virtualized env[1]
- latest overcloud-hardened-uefi-full.qcow2 image are successfully booting on that env, meaning the OC is successfully provisioned and ready to deploy

In your case, I'd suggest to ensure the tftp server serving the pxe things is reachable from the node. If it's not the case, it would mean some weird network issue, or that the ironic part used to get the tftp up is broken at some point. Though this latter possibility is, probably, wrong, seeing I could deploy the whole thing today (using tripleo-ci-testing content).

[1] https://github.com/cjeanner/tripleo-lab

[Dariusz Smigiel (smigiel-dariusz)]

> 20:27 <guilhermesp_> hum im wondering which instance types are being used, maybe the server are going to another aggregate that doesnt have uefi enabled ?

[Guilherme Steinmuller Pimentel (guilhermesp)]

I checked the hypervisors on the ci aggregate and it looks like everything looks good in there.

I was able to boot a UEFI vm on them by using a ubuntu image which was built using dib, with the following elements:

dib elements:

~/dib$ cat ubuntu-uefi-grub2.d/dib-manifests/dib_arguments
ubuntu vm block-device-efi dhcp-all-interfaces grub2 -o ubuntu-uefi-grub2

VM result:

root@test:~# ls /sys/firmware/efi
config_table efivars fw_platform_size fw_vendor runtime runtime-map systab vars
root@test:~#
root@test:~#
root@test:~# test -d /sys/firmware/efi && echo efi || echo bios
efi

[Dariusz Smigiel (smigiel-dariusz)]

Steve Baker identified the issue: https://review.opendev.org/c/openstack/openstack-virtual-baremetal/+/834813

[Dariusz Smigiel (smigiel-dariusz)]

The change has been merged.