Adjust for Cluster Architecture

There are three architecture options:

  • SNO

  • Compact

  • Full

SNO has some clear limitations and sizing is relatively straightforward, so we can ignore it for this exercise.

Compact and full are two sides of the same coin, but we need to know other info too:

  • Does the customer want homogenous node sizes or are they willing to use a different node size for control plane and infra?

    • If yes, then consider having schedulable control plane nodes that also host infra workload. Depending on the node size/capacity, there may still be capacity for customer workload on these nodes.

  • Will this cluster be using hosted control planes, where there are no control plane nodes?

  • Which of the infra qualified add-on features, e.g. logging, etc., will be hosted by the cluster?

    • If the cluster does not have dedicated infra (or combined infra + control plane) nodes, this capacity will need to be added to the overall requirement.

  • Is the customer planning on using features, such as pipelines and/or GitOps, which will increase the API server, and other control plane component, utilization?

    • Each of these places additional burden on the control plane components. This may affect the size of those nodes if they’re dedicated or how many VMs can be hosted on the control plane nodes.

Let’s say that the customer has chosen the 11-node cluster size from previous section. They want to have schedulable control plane nodes that are able to host their workloads, multiple additional OpenShift Platform Plus features, and intend to use homogeneous nodes. For the sake of simplicity, let’s assume that the combined resource requirements for this totals approximately 15% of the three control plane nodes. This means that the cluster needs to have and additional (3 * 15) / (11 * 100) = 4% total cluster capacity for non-VM workloads. Since we do not already have this much excess capacity available (we rounded up from 10.4 → 11 nodes, which gives us 60% of one node distributed across the cluster, however that capacity is consumed by overhead), this would require a 12th node to be added to the cluster.