Chapter 7. How to use dedicated worker nodes for Red Hat OpenShift Container Storage

Using infrastructure nodes to schedule Red Hat OpenShift Container Storage resources saves on Red Hat OpenShift Container Platform subscription costs. Any Red Hat OpenShift Container Platform (RHOCP) node that has an infra node-role label requires an OpenShift Container Storage subscription, but not an RHOCP subscription.

It is important to maintain consistency across environments with or without Machine API support. Because of this, it is highly recommended in all cases to have a special category of nodes labeled as either worker or infra or have both roles. See the Section 7.3, “Manual creation of infrastructure nodes” section for more information.

7.1. Anatomy of an Infrastructure node

Infrastructure nodes for use with OpenShift Container Storage have a few attributes. The infra node-role label is required to ensure the node does not consume RHOCP entitlements. The infra node-role label is responsible for ensuring only OpenShift Container Storage entitlements are necessary for the nodes running OpenShift Container Storage.

  • Labeled with node-role.kubernetes.io/infra

Adding an OpenShift Container Storage taint with a NoSchedule effect is also required so that the infra node will only schedule OpenShift Container Storage resources.

  • Tainted with node.ocs.openshift.io/storage="true"

The label identifies the RHOCP node as an infra node so that RHOCP subscription cost is not applied. The taint prevents non OpenShift Container Storage resources to be scheduled on the tainted nodes.

Example of the taint and labels required on infrastructure node that will be used to run OpenShift Container Storage services:

    spec:
      taints:
      - effect: NoSchedule
        key: node.ocs.openshift.io/storage
        value: "true"
      metadata:
        creationTimestamp: null
        labels:
          node-role.kubernetes.io/worker: ""
          node-role.kubernetes.io/infra: ""
          cluster.ocs.openshift.io/openshift-storage: ""

7.2. Machine sets for creating Infrastructure nodes

If the Machine API is supported in the environment, then labels should be added to the templates for the Machine Sets that will be provisioning the infrastructure nodes. Avoid the anti-pattern of adding labels manually to nodes created by the machine API. Doing so is analogous to adding labels to pods created by a deployment. In both cases, when the pod/node fails, the replacement pod/node will not have the appropriate labels.

Note

In EC2 environments, you will need three machine sets, each configured to provision infrastructure nodes in a distinct availability zone (such as us-east-2a, us-east-2b, us-east-2c). Currently, OpenShift Container Storage does not support deploying in more than three availability zones.

The following Machine Set template example creates nodes with the appropriate taint and labels required for infrastructure nodes. This will be used to run OpenShift Container Storage services.

  template:
    metadata:
      creationTimestamp: null
      labels:
        machine.openshift.io/cluster-api-cluster: kb-s25vf
        machine.openshift.io/cluster-api-machine-role: worker
        machine.openshift.io/cluster-api-machine-type: worker
        machine.openshift.io/cluster-api-machineset: kb-s25vf-infra-us-west-2a
    spec:
      taints:
      - effect: NoSchedule
        key: node.ocs.openshift.io/storage
        value: "true"
      metadata:
        creationTimestamp: null
        labels:
          node-role.kubernetes.io/infra: ""
          cluster.ocs.openshift.io/openshift-storage: ""
Important

If you add a taint to the infrastructure nodes, you also need to add tolerations to the taint for other workloads, for example, the fluentd pods. For more information, see the Red Hat Knowledgebase solution Infrastructure Nodes in OpenShift 4.

7.3. Manual creation of infrastructure nodes

Only when the Machine API is not supported in the environment should labels be directly applied to nodes. Manual creation requires that at least 3 RHOCP worker nodes are available to schedule OpenShift Container Storage services, and that these nodes have sufficient CPU and memory resources. To avoid the RHOCP subscription cost, the following is required:

oc label node <node> node-role.kubernetes.io/infra=""
oc label node <node> cluster.ocs.openshift.io/openshift-storage=""

Adding a NoSchedule OpenShift Container Storage taint is also required so that the infra node will only schedule OpenShift Container Storage resources and repel any other non-OpenShift Container Storage workloads.

oc adm taint node <node> node.ocs.openshift.io/storage="true":NoSchedule
Warning

Do not remove the node-role node-role.kubernetes.io/worker=""

The removal of the node-role.kubernetes.io/worker="" can cause issues unless changes are made both to the OpenShift scheduler and to MachineConfig resources.

If already removed, it should be added again to each infra node. Adding node-role node-role.kubernetes.io/infra="" and OpenShift Container Storage taint is sufficient to conform to entitlement exemption requirements.