Hello Mike, We had this issue in the past and there was a bug tracker upstream [1], the upstream fix was merged during xena cycle [2]. I would attempt to check the firmware present in the machine and try to upgrade to see if it helps. I've also noticed we just had another change in IPA with a fix for efibootmgr [3], not sure if the image you downloaded from tarballs already contains. [1] https://storyboard.openstack.org/#!/story/2008962 [2] https://review.opendev.org/c/openstack/ironic-python-agent/+/795862 [3] https://review.opendev.org/c/openstack/ironic-python-agent/+/881762 Em qui., 25 de mai. de 2023 às 04:19, Mike Currin <mike@idia.ac.za> escreveu:
Hi All,
We have a Xena based Openstack deployment, recently we deployed 60+ nodes in our research cluster with Ironic which worked well. All of these were deployed using a standard process I'll describe below.
We recently took delivery of a new Dell R6625 server with NVMe devices onlym which only support UEFI boot, so we are trying to get that working.
The server PXEs and downloads the RAM disk and then the Deploy image, once running that it immediately crashes (I assume when running linuxefi). We tested UEFI deploy on an existing Dell R640 server, that server works with BIOS but we swapped it over to UEFI and it does the same, so it wasn't due to the much bigger/different architecture (AMD vs Intel) server. We have a few older servers in a test setup (which are Dell R630's) which are working fine and don't do this behaviour. We haven't tried them on our production setup as if even if they worked it wouldn't help us move forward.
I made a video showing this: https://www.dropbox.com/s/5jbn1qpylxaevqb/uefiboot2.mov?dl=0 In the iDRAC we just get that the "System BIOS has halted" and somewhere I said to change hardware that you recently added, which feels unlikely as 2 different servers both working elsewhere with totally different hardware,.
I've done a iDRAC Serial console debug but it isn't showing me much that is of any use.
This is our entire process to deploy a node (some is once off of course, I've not included the network setup):
openstack flavor create --ram 256000 --disk 20 --vcpus 32 --public our-baremetal openstack flavor set our-baremetal --property capabilities:boot_mode="uefi"
We downloaded the latest (Xena) images from: https://tarballs.opendev.org/openstack/ironic-python-agent/dib/files/ I also tried the latest Centos9 ones just to try them, made no difference.
We then extract the image and make a small mod to the service start (we found it didn't bring the NIC immediately up so put a ping delay in the ExecStart), but that's not part of this problem.
openstack image create --disk-format aki --container-format aki --public --file ironic-agent.kernel deploy-vmlinuz openstack image create --disk-format ari --container-format ari --public --file ironic-agent.initramfs-ping-patched deploy-initrd
Then to do the deploy: export HOSTNAME=<hostname> export MGMTIP=<idracip>
openstack baremetal node create --driver ipmi --name $HOSTNAME --driver-info ipmi_port=623 --driver-info ipmi_username=root --driver-info 'ipmi_password=<ourpassword>' --driver-info ipmi_address=$MGMTIP --resource-class baremetal-resource-class --property cpus=32 --property memory_mb=256000 --property local_gb=20 --property cpu_arch=x86_64 --driver-info deploy_ramdisk=$(openstack image show deploy-initrd -f value -c id) --driver-info deploy_kernel=$(openstack image show deploy-vmlinuz -f value -c id) NODE=$(openstack baremetal node show -f value -c uuid $HOSTNAME) openstack baremetal node set $NODE --property capabilities='boot_mode:uefi'
openstack baremetal port create <MACADDRESS> --node $NODE --physical-network physnet3 openstack baremetal node manage $NODE --wait && openstack baremetal node list && openstack baremetal node provide $NODE && openstack baremetal node list
openstack server create --use-config-drive --image <ourimage> --flavor our-baremetal --security-group worker --network ironic-network --key-name <ourkeyname> servername
Does any one have any more info to help or any suggestions as to something more I could try, I'm out of ideas. I know that UEFI itself works on both the servers, we have a setup with Ubuntu MAAS and it can deploy perfectly fine using its process with the UEFI setup so, it's something on the Ironic deploy image that's causing us this problem.
Regards, Mike
-- *Att[]'s* *Iury Gregory Melo Ferreira * *MSc in Computer Science at UFCG* *Ironic PTL * *Senior Software Engineer at Red Hat Brazil* *Social*: https://www.linkedin.com/in/iurygregory *E-mail: iurygregory@gmail.com <iurygregory@gmail.com>*