Typically, the PodIP for a pod using host networking is set to the node's IP. The node IP is on the baremetal network (192.168.111.x). However, the container runtime can override that, and a bug in CRI-O is causing this to happen, and it was picking the IP on a provisioning interface: [core@master-0 ~]$ sudo crictl inspectp 1a8c9a6d3d05a { "status": { "id": "1a8c9a6d3d05a162340d018cec7146d8881987784afce607f9c0d277519a65bf", "metadata": { "attempt": 0, "name": "kube-apiserver-master-0", "namespace": "openshift-kube-apiserver", "uid": "d842b9a1988d8656c555a333e12c2200" }, "state": "SANDBOX_READY", "createdAt": "2019-11-12T00:15:47.881389131Z", "network": { "ip": "172.22.0.90" }, This is fixed in cri-o/cri-o#2944 and should restore the previous behavior. This fix is present in 43.81.201911131545.0.
From https://github.com/openshift/installer/blob/release-4.3/data/data/rhcos.json, boot image rhcos version is bumped to "rhcos-43.81.201911192044", so move this bug to verified. BTW, baremetal ipi install is not in OCP QE test scope, if this issue still reproduced, pls reopen.
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2020:0062