How Do I Configure the Add-on Resource Quotas Based on Cluster Scale?

Configuring Resource Quotas for coredns

Queries per Second (QPS) of the coredns add-on is positively correlated with the CPU consumption. If the number of nodes or containers in the cluster grows, the coredns pod will bear heavier workloads. Adjust the number of add-on pods and their CPU and memory quotas based on the cluster scale.

Configuring Resource Quotas for everest

After the cluster scale is adjusted, the everest specifications need to be modified based on the cluster scale and the number of PVCs. The requested CPU and memory can be increased based on the number of nodes and PVCs. For details, see Table 2.

In non-typical scenarios, the formulas for estimating the limit values are as follows:

• everest-csi-controller
  • CPU limit: 250m for 200 or fewer nodes, 350m for 1000 nodes, and 500m for 2000 nodes
  • Memory limit = (200 MiB + Number of nodes x 1 MiB + Number of PVCs x 0.2 MiB) x 1.2
• everest-csi-driver
  • CPU limit: 300m for 200 or fewer nodes, 500m for 1000 nodes, and 800m for 2000 nodes
  • Memory limit: 300 MiB for 200 or fewer nodes, 600 MiB for 1000 nodes, and 900 MiB for 2000 nodes

Configuring Resource Quotas for autoscaler

autoscaler automatically adjusts the number of nodes in a cluster based on workloads. Adjust the number of add-on pods and their CPU and memory quotas based on the cluster scale.

Configuring Resource Quotas for volcano

After the cluster scale is increased, the resource quotas required by volcano need to be modified based on the cluster scale.

• If the number of nodes is less than 100, retain the default configuration. The requested CPU is 500m, and the limit is 2000m. The requested memory is 500 MiB, and the limit is 2000 MiB.
• If the number of nodes is greater than 100, increase the requested CPU by 500m and the requested memory by 1000 MiB each time 100 nodes (10,000 pods) are added. Increase the CPU limit by 1500m and the memory limit by 1000 MiB.
  

  Formulas for calculating the requests:

  • CPU request: Calculate the number of nodes multiplied by the number of pods, perform interpolation search using the product of the number of nodes in the cluster multiplied by the number of pods in Table 4, and round up the request and limit that are closest to the specifications.

    For example, for 2000 nodes (20,000 pods), the product of the number of nodes multiplied by the number of pods is 40 million, which is close to 700/70,000 in the specification (Number of nodes x Number of pods = 49 million). Set the CPU request to 4000m and the limit to 5500m.

  • Memory request: Allocate 2.4 GiB of memory to every 1000 nodes and 1 GiB of memory to every 10,000 pods. The memory request is the sum of the two values. (The obtained value may be different from the recommended value in Table 4. You can use either of them.)

    Memory request = Number of nodes/1000 x 2.4 GiB + Number of pods/10000 x 1 GiB

    For example, for 2000 nodes and 20,000 pods, the memory request value is 6.8 GiB (2000/1000 x 2.4 GiB + 20000/10000 x 1 GiB).

Configuring Resource Quotas for Other Add-ons

Resource quotas of other add-ons may also be insufficient due to cluster scale expansion. If, for example, the CPU or memory usage of the add-on pods increases and even OOM occurs, modify the resource quotas as required.

For example, the resources occupied by the kube-prometheus-stack add-on are related to the number of pods in the cluster. If the cluster scale is expanded, the number of pods may also grow. In this case, increase the resource quotas of the prometheus pods.