We currently build this list using a pdb query that returns all hosts with the Profile::Kubernetes::Mediawiki_runner declared, including nodes in Failed state in netbox.
While this is not an operational issue as the node would download the image automatically when back in a ready state and scheduled for a mediawiki pod, it is causing confusion for the deployers who see an error at the docker_pull_k8s stage.
Switching to using the cluster=kubernetes,service=kubesvc conftool query would allow for the depooling of nodes to exclude them from the list, removing the confusing errors. In order to avoid pulling the image on sessionstore nodes, we can add a check in mediawiki-image-download.sh.