We previously ran into issues running an IBU from X to Y, where X and Y have the same rhcos base image (OCPBUGS-25951). A workaround was found to resolve the issue in this scenario. However, an additional issue has been found when running consecutive IBUs from X to Y and then Y to Z, where Y and Z have the same rhcos base image.

In the Y to Z upgrade, the pull-local and deploy commands are successful, with the workaround in place from OCPBUGS-25951. However, once we run the ostree admin set-default command, we get the following error:

loading sysroot: Parsing deployment /ostree/boot.1/rhcos_4.15.0_rc.8/827728895611cec1dd1d51a4e88e64d96a0dbf0fbcb8dbd0d85e5d9634691ff8/0 in stateroot rhcos_4.15.0_rc.8: readlinkat: No such file or directory

The Y and Z deployments have the same base rhcos image, same kernel, and the same bootloader config options, but are in separate stateroots. The deployment_bootconfigs_equal function in ostree-sysroot-deploy.c, however, doesn't look at the stateroot, so it returns TRUE when comparing them. When the deployments are swapped by the set-default command, the bootserials are modified and new symlinks are created. But since deployment_bootconfigs_equal returns true, the ostree_sysroot_write_deployments_with_options function does not update the bootloader config. After this point, ostree is unable to parse the deployment references from the bootloader config files as they are stale.

As a workaround, LCA has been updated to include an unused karg when running the deploy command (ibu=$VERSION). This ensures that the boot config options will be unique between deployments, so that set-default updates the bootloader config.

    1.
    2.
    3.