Fuzzball Documentation
Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Back to homepage

Remaining Dependencies

Fuzzball requires a few more components to be present and operational before using the operator for installation.

In a 3-node HA setup, the second and third nodes need only the NFS packages. See below.

metallb

A bare-metal deployment of RKE2 requires the metallb load-balancer for the assignment of external IP addresses.

# kubectl apply -f \
    https://raw.githubusercontent.com/metallb/metallb/v0.15.2/config/manifests/metallb-native.yaml

Once metallb is deployed, add a pool to tell K8s what addresses it is allowed to use for external services. Adjust the address list for the local environment. Fuzzball will require an address in addition to the address used for the prerequisite OCI registry.

# IP1="" # set this according to your environment and preference (for example 10.0.0.99)

# IP2="" # set this according to your environment and preference (for example 10.0.0.100)
# cat >metallb-pool.yaml<<EOF
apiVersion: metallb.io/v1beta1
kind: IPAddressPool
metadata:
  name: default-pool
  namespace: metallb-system
spec:
  addresses:
  - ${IP1}/32
  - ${IP2}/32
EOF

# kubectl apply -f metallb-pool.yaml

After applying the pool to the cluster, create an advertisement to make the pool available on a local L2 network.

# INTERNAL_INTERFACE="" # populate this with the value of your internal interface (e.g. enp8s0)
# cat >metallb-advertisement.yaml<<EOF
apiVersion: metallb.io/v1beta1
kind: L2Advertisement
metadata:
  name: default-advertisement
  namespace: metallb-system
spec:
  ipAddressPools:
  - default-pool
  interfaces:
  - ${INTERNAL_INTERFACE}
EOF

# kubectl apply -f metallb-advertisement.yaml

Helm

Many Kubernetes applications (including the Fuzzball Operator) use the Helm “package manager” to manage deployments. The official Helm installation instructions recommend a curl | sh approach.

It is usually considered a bad practice to perform a curl | sh since you never can be completely sure of the code you are running. A bad actor could compromise the server hosting the URL and change the script. You may want to download and inspect the script before running it.
# curl -sfL https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash -

StorageClass

The Fuzzball Operator expects a StorageClass to be deployed as a prerequisite. If you are installing Fuzzball on a single server node, we recommend you use Local Path Provisioner. If you are installing Fuzzball in a 3-node HA setup, we suggest you use the Longhorn distributed block storage system.

Remember which StorageClass you install. Unless it is the cluster’s default StorageClass, pass it to fuzzball cluster kubernetes deploy with --storage-class.

Please select the tab with instructions appropriate for your environment.

Please select either the single node (Local Path Provisioner) or HA (Longhorn) tab for instructions appropriate to your environment.

Install Local Path Provisioner with the following command (or similar).

# kubectl apply -f \
    https://raw.githubusercontent.com/rancher/local-path-provisioner/v0.0.28/deploy/local-path-storage.yaml

RKE2 ships with an SELinux policy that permits container access to /opt/local-path-provisioner when it carries the container_file_t label. To ensure that the path is created and properly labeled, run the following.

# mkdir -p /opt/local-path-provisioner

# restorecon -vr /opt/local-path-provisioner

ls -Zd /opt/local-path-provisioner must then show the type container_file_t. restorecon sets it with container-selinux 2.211 or later. With an older version, add the rule first, then run restorecon again:

# semanage fcontext -a -t container_file_t '/opt/local-path-provisioner(/.*)?'

Instead of installing Local Path Provisioner here, you can let fuzzball cluster kubernetes deploy --install-storage-class local-path install it as the default StorageClass. On nodes with SELinux in enforcing mode, still create and label the directory as above on every node first: until you do, preflight fails and prints the same command.

Install the Longhorn distributed block storage system with the following commands.

# systemctl enable --now iscsid

# kubectl apply -f https://raw.githubusercontent.com/longhorn/longhorn/v1.10.1/deploy/longhorn.yaml

NFS

Fuzzball Orchestrate uses NFS to deliver configuration to the compute nodes and share containers. This shared filesystem can easily be provided by NFS from the Server node, particularly in single-node deployments.

Every K8s node needs the NFS client tools, because the Fuzzball Operator mounts volumes from the NFS share on the K8s nodes. Install them on the Server node even if your NFS server is a separate machine (use nfs-common instead of nfs-utils on Debian or Ubuntu):

# dnf install -y nfs-utils

If you want to serve NFS from the Server node, follow these steps as well:

# PRIVATE_SUBNET="" # populate this with the proper value for your environment (e.g. 10.0.0.0/20)
# systemctl enable --now nfs-server

# mkdir -p /srv/fuzzball/shared

# echo "/srv/fuzzball/shared ${PRIVATE_SUBNET}(rw,sync,no_subtree_check,no_root_squash)" >>/etc/exports

# exportfs -a

In a 3-node HA setup, it is necessary to install the NFS packages on the second and third node. It is not required to mount /srv/fuzzball/shared from the first node, or carry out any of the other steps detailed above on nodes 2 and 3.

Run the following command on your remaining server nodes.

# dnf install -y nfs-utils

Congratulations! At this point you should have a working K8s (RKE2) deployment suitable for a Fuzzball installation. If you decide that you want to remove the deployment, run the rke2-uninstall.sh script that the RKE2 installer placed on the server.

Now head over to the Fuzzball Deployment Guide to complete your setup.