Control Plane Node Scaling
The control plane node resource requirements depent on a number of different metrics. CPU requirements generally depend on the ammount on API write requests, while memory requirements generally depend on the amount of resource objects (most importantly Pod count) in the cluster.
To avoid cascading failures during a scheduled node reboot or an unexpected node failure, it’s recommended to keep the overall CPU and memory usage not significantly higher than 60%.
Guidelines
The following guidelines are an estimate, based on metrics collected from our managed clusters. Individual clusters may vary depending on worker node size and individual workload running on the cluster.
|
Many cloud providers don’t support arbitrary VM sizes, so we’ll usually recommend the closest available VM size that’s at least as large as the recommended size below. |
| Control Plane Spec | Cluster Parameters |
|---|---|
4 CPU 16Gi Memory |
Small Cluster max. 3 Worker Nodes |
4 CPU 24Gi Memory |
Small Cluster more than 3 Worker Nodes more than 500 Pods |
4 CPU 32Gi Memory |
Medium Cluster more than 1000 Pods |
References
See the capacity dashboards (VSHN internal link) for more details about the current cluster sizes.