Permanently removing a Cloudera Embedded Container Service host

You can permanently remove a Cloudera Embedded Container Service host.

On the host detail page in Cloudera Manager, click Actions → Stop Roles on Host.
On the host detail page, select the ECS role and click Delete.
On the same page, select the Docker role and click Delete.
Step 4 (On-host cleanup)
1. Log in to the host as root (or using sudo -i) and run the following sequence: rke2-killall.sh is not fully idempotent when containerd-shim children respawn — running it twice is normally sufficient.cd /opt/cloudera/parcels/ECS/bin
  ./rke2-killall.sh # usually 2 times is sufficient ./rke2-killall.sh
2. Unmount all NFS mounts on the host before running the uninstaller — the RKE2 uninstaller does not unmount NFS:
```
# List NFS mounts, then umount each 
mount -t nfs,nfs4 
umount -a -t nfs,nfs4
```
3. Run the uninstaller and clear Cloudera Embedded Container Service-specific data directories:
```
./rke2-uninstall.sh
rm -rf /ecs/*                     # assumes the default defaultDataPath and lsoDataPath
rm -rf /var/lib/docker_server/*   # deletes the auth and certs
rm -rf /etc/docker/certs.d/*      # deletes the ca.crt
rm -rf /docker                    # assumes the default defaultDataPath for docker
rm -rf /var/lib/rancher/*
```
  If any rm -rf reports Device or resource busy, re-run ./rke2-killall.sh and check for surviving processes:
```
ps -ef | grep -E 'containerd-shim|kubelet|rke2' | grep -v grep
```
Navigate to Cloudera Manager → Hosts → All Hosts, select the host, and choose Actions for Selected → Remove From Cloudera Manager.
Review the following notes:
- rke2-uninstall.sh cleans only /var/lib/rancher/rke2, /etc/rancher/rke2, /etc/rancher/node, /etc/cni, /opt/cni/bin, and /var/lib/kubelet. Every other path in Step d (On-host cleanup) must be removed manually or the host will retain stale Cloudera Embedded Container Service state on re-add.
- rke2-killall.sh unmounts bind mounts under the kubelet root but does not touch NFS mounts. Skip the explicit umount and the uninstaller can hang on stale NFS handles.
- If the host used a non-default defaultDataPath or lsoDataPath, replace /ecs/* and /docker in the cleanup with the configured paths from Cloudera Manager. The values are visible under the Cloudera Embedded Container Service service configuration.
Rollback options:
- Recommission: If the host still misbehaves after End Maintenance, place it back into maintenance mode from the same Actions menu. No data is lost.
- Permanent removal: Rollback is not possible once Step 4 (On-host cleanup) is initiated. Restoring the host to the cluster requires the standard Cloudera Embedded Container Service add-host workflow — parcel re-install, ECS/Docker role assignment, and Longhorn re-onboarding of the new node.