Hi,

Description of problem

Bare Metal IPI provisioning is failing to provision the worker nodes. The metal3-machine-os-downloader InitContainer is getting in CrashLoopBackOff state because it cannot find virt-* commands in the container image.

> oc -n openshift-machine-api get pods | grep -v Running
NAME                       READY   STATUS
metal3-fc66f5846-gtq9m     0/7     Init:CrashLoopBackOff
metal3-image-cache-d4qcz   0/1     Init:1/2
metal3-image-cache-djzcf   0/1     Init:1/2
metal3-image-cache-p5mwg   0/1     Init:1/2

> oc -n openshift-machine-api logs deployment/metal3 -c metal3-machine-os-downloader
[omitted]
++ LIBGUESTFS_BACKEND=direct
++ virt-filesystems -a rhcos-412.86.202207142104-0-openstack.x86_64.qcow2 -l
/usr/local/bin/get-resource.sh: line 88: virt-filesystems: command not found
++ grep boot
++ cut -f1 '-d '
+ BOOT_DISK=
++ LIBGUESTFS_BACKEND=direct
++ virt-ls -a rhcos-412.86.202207142104-0-openstack.x86_64.qcow2 -m '' /boot/loader/entries
/usr/local/bin/get-resource.sh: line 90: virt-ls: command not found
+ BOOT_ENTRIES=
+ rm -fr /shared/tmp/tmp.CnCd2E3kxN

Version-Release number of selected component (if applicable):

OpenShift 4.12.0-ec.0+

Analysis

Since https://github.com/openshift/ocp-build-data/pull/1757, the ironic-machine-os-downloader container image is built using RHEL9 repositories.

However, following upstream move of guestfs tools to a dedicated repository [1], the libguestfs packaging differs between RHEL8 and RHEL9:

• the libguestfs-tools-c package containing most virt-* commands is now provided by the guestfs-tools package
• the libguestfs-tools package is now provided by the virt-win-reg package which does not require the libguestfs-tools-c package anymore

Since the Dockerfile specifies only the libguestfs-tools package, the virt-* commands are not installed when using RHEL9 repositories.

A trivial fix is to update the Dockerfile to install the guestfs-tools package instead of the libguestfs-tools package.

Regards,

Denis