It would be nice if this were more consistent with infrastructure nodes. The docs on creating infra machinesets recommend using node-role.kubernetes.io/infra='' to identify those. According to a knowledge base article, cloud.redhat.com uses the same label to identify infra nodes and exclude them from reported license counts.

One could use node-role.kubernetes.io/storage='' for dedicated storage nodes, or apply both role.kubernetes.io/storage='' and role.kubernetes.io/worker='' to mixed nodes. That could make it easier for cloud.redhat.com to display accurate license counts as well, for both worker and storage nodes.

It would also let the admin avoid delving into taints and tolerations, which this document doesn't adequately describe. For example neither this document nor the provided link to https://access.redhat.com/documentation/en-us/openshift_container_platform/4.2/html-single/nodes/index#nodes-scheduler-taints-tolerations doesn't explain how to update the ClusterLogging custom resource to apply the toleration to fluentd, or even mention the need to do so.

Someone implementing OCS for the first time shouldn't have to troubleshoot the application of tolerations to operator-managed pods. Ideally the recommendation would be consistent with infra nodes and use the node-role. If there's a real need for taints and tolerations, then there should be more thorough guidance on how to correctly apply tolerations to the components that are mentioned (fluentd, node-ca, and node-exporter).

Reported by: chosey

https://access.redhat.com/documentation/en-us/red_hat_openshift_container_storage/4.2/html/deploying_openshift_container_storage/deploying-openshift-container-storage#annotations:03413c1c-ee8b-4a7b-8726-2cca321821e5