-
Bug
-
Resolution: Unresolved
-
Major
-
None
-
4.14.z, 4.16.z
-
Important
-
None
-
False
-
Description of problem:
During installation of single-node OpenShift with IBM Power, encountering the disk busy error while the installer tries to run install-to-disk.sh to apply the master.ign file to the install disk.
Version-Release number of selected component (if applicable):
OCP 4.14
How reproducible:
Always
Steps to Reproduce:
1. Try to install ocp 4.14 SNO on IBM Power 9 LPAR 2. Used only one disk for cluster installation on the node 3. Specify the installation disk under bootstrapInPlace in install-config YAML bootstrapInPlace: installationDisk: /dev/disk/by-id/wwn-0x600507680c818022f00000000000f21e
Actual results:
Oct 08 20:01:09 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162016]: Executing coreos-installer with the following options: install -i /opt/openshift/master.ign /dev/disk/by-id/wwn-0x600507680c818022f00000000000f21e Oct 08 20:01:09 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162046]: Downloading Fedora CoreOS stable ppc64le metal image (raw.xz) and signature Oct 08 20:01:10 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162046]: Partitions in use on /dev/disk/by-id/wwn-0x600507680c818022f00000000000f21e: Oct 08 20:01:10 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162046]: /dev/sdb3 mounted on /boot Oct 08 20:01:10 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162046]: Error: checking for exclusive access to /dev/disk/by-id/wwn-0x600507680c818022f00000000000f21e Oct 08 20:01:10 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162046]: Caused by: Oct 08 20:01:10 xxx-master.pxxx.xxx.xxx.com install-to-disk.sh[1162046]: found busy partitions
Expected results:
Successful installation of sno
Additional info:
As per customer, they have attempted the installation on both IBM Power and IBM/Z LPARs, and they are not using multipathing. They also performed the installation on an x86 KVM guest with a single attached virtio disk, and they confirm there is no multipathing involved in this setup either. For this reason, we can rule out multipathing as a contributing factor to the issue we are encountering. Across the different installation platforms, we consistently reach the same point of failure: when the master ignition file is applied by the coreos-installer. I believe that the installation is failing because the same disk used for the bootstrap process is also intended for the single OpenShift master node.