Failed snapshot restore for RKE2 clusters with misplaced mountpoint for the RKE2 binaries/config files

A standalone or Rancher-provisioned RKE2 cluster, in which a rollback is performed from: 

• a v1.22 patch version >= v1.22.16+rke2r1, 
• a v1.23 patch version >= v1.23.15+rke2r1,
• a v1.24 patch version >= v1.24.9,
• a v1.25 patch version >= v1.25.5+rke2r1,
• or any RKE2 release >= v1.26.0+rke2r1

to:

• a v1.22 patch version < v1.22.16+rke2r1,
• a v1.23 patch version < v1.23.15+rke2r1,
• a v1.24 patch version < v1.24.9,
• a v1.25 patch version < v1.25.5+rke2r1,
• or any RKE2 release <= v1.21.14+rke2r1

The etcd snapshot restore will fail.

Workaround:

Execute on all nodes of the cluster before initiating the restore.

rm /etc/systemd/system/rke2-*.service
systemctl daemon-reload

• The RKE2 install script for v1.24.4 does not install to /opt/rke2 if /usr/local is on a separate mount.
• The RKE2 install script for v1.24.17 does install to /opt/rke2 if /usr/local is on a separate mount.
• As part of the copy process, the systemd unit files are stored into an override in /etc/system/systemd, with /opt/rke2 defined in the exec path
• When you rollback from v1.24.17 to v1.24.4, the binaries are written to /usr/local, but the systemd unit file override remains in place pointing to /opt/rke2 and the v1.24.17 binaries. The restore will fail as a result of not finding the right binaries and systemd config files.
• This behaviour is the result of an earlier bug in the RKE2 install script that was fixed in the following script.