To be able to run fully offline, MetalK8s comes with its own registry serving all necessary images used by its containers. This registry container sits on the Bootstrap node.
With a highly available registry, container images are served by multiple nodes, which means the Bootstrap node can be lost without impacting the cluster. It allows pods to be scheduled, even if the needed images are not cached locally.
This procedure only talk about registry HA as Bootstrap HA is not supported for the moment, so it’s only a part of the Bootstrap functionnaly. Check this ticket for more informations https://github.com/scality/metalk8s/issues/2002
Prepare the node¶
To configure a node to host a registry, a
repository pod must be scheduled
This node must be part of the MetalK8s cluster and no specific roles or
taints are needed.
All ISOs listed in the
archives section of
must be copied from the Bootstrap node to the target node at exactly the same
Deploy the registry¶
Connect to the node where you want to deploy the registry and run the following salt states
Prepare all the MetalK8s ISOs
root@node-1 $ salt-call state.sls \ metalk8s.archives.mounted \ saltenv=metalk8s-124.1.7-dev
If you have some solutions, prepare the solutions ISOs
root@node-1 $ salt-call state.sls \ metalk8s.solutions.available \ saltenv=metalk8s-124.1.7-dev
Deploy the registry container
root@node-1 $ salt-call state.sls \ metalk8s.repo.installed \ saltenv=metalk8s-124.1.7-dev
Reconfigure the container engines¶
Containerd must be reconfigured to add the freshly deployed registry to its endpoints and so it can still pull images in case the Bootstrap node’s one is down.
From the Bootstrap node, run (replace
<bootstrap_node_name> with the
actual Bootstrap node name):
root@bootstrap $ kubectl exec -n kube-system -c salt-master \ --kubeconfig=/etc/kubernetes/admin.conf \ salt-master-<bootstrap_node_name> -- salt '*' state.sls \ metalk8s.container-engine saltenv=metalk8s-124.1.7-dev