Scaling down the cluster
Scaling down the cluster refers to the removal of cluster members.
These instructions show how to remove a node with the exception of the primary (bootstrap) node. For instructions on the latter, see page Removing the primary node. Note that, in terms of removing nodes, the primary node is always the last to be removed.
Remove the node
On the primary node:
sunbeam cluster remove --name <node FQDN>
Note: A current software issue (fixed in the edge
risk level) causes the cluster remove
command to fail with a message like:
ERROR removing machine failed: machine 1 has unit "sunbeam-machine/1" assigned
.
To work around this manually remove the unit in the error message by issuing:
juju remove-unit sunbeam-machine/1
The cluster remove
command can then be reissued.
Repeat the workaround as needed.
Remove components from the node
Software components now need to removed from the target node. Perform all the below steps on the target node.
Remove the Juju agent:
sudo /sbin/remove-juju-services
Remove the juju snap:
sudo snap remove --purge juju
Remove Juju configuration:
rm -rf ~/.local/share/juju
Remove the openstack-hypervisor and openstack snaps:
sudo snap remove --purge openstack-hypervisor
sudo snap remove --purge openstack
Remove openstack snap configuration:
rm -rf ~/.local/share/openstack
Remove the k8s snap:
sudo k8s remove-node
sudo snap remove --purge k8s
The above steps can take a few minutes to complete.
Remove the disk(s) used by microceph on this node:
sudo microceph disk list
sudo microceph disk remove <OSD on this node>
Remove the microceph snap:
sudo snap remove --purge microceph
If required clean the disk(s) identified in the earlier command:
sudo dd if=/dev/zero of=<DISK PATH> bs=4M count=10
Caution: The dd
command will result in the permanent erasure of data. It is vital that you have specified the correct disk path to avoid unintended data loss.
Clear the remaining network configuration with a reboot:
sudo reboot