Evaluating Resources for Global Cluster

Overview

Use these resource evaluation practices when planning the global cluster. The sizing depends on how the global cluster is used:

  • In a Multi-Cluster deployment, the global cluster runs only the platform. Select the tier by the number of managed clusters; see Node Sizing.
  • In a Single Cluster deployment, the global cluster also serves as the workload cluster. Select the tier by the number of worker nodes; see Global Cluster as a Workload Cluster.

Proper node sizing helps the global cluster manage registered clusters, handle synchronization traffic, and process user API and Web Console requests within the performance expectations of your environment.

Node Sizing

This section applies to the global cluster of a Multi-Cluster deployment.

The global cluster is responsible for:

  • Maintaining cluster registration and metadata.
  • Handling inbound API requests from the Web Console and CLI.
  • Coordinating synchronization and heartbeat messages with connected clusters.
  • Managing internal controllers and resource reconciliation loops.

Because the global cluster handles both management operations and data aggregation from all connected clusters, plan resource allocation according to the expected scale and workload intensity.

Baseline Production Sizing

The production-scale sizing depends primarily on:

  • Number of managed clusters
  • Frequency of synchronization cycles
  • Concurrent API request rate (from users or automation)
  • Volume of streaming requests
  • Number of plugins installed

The following table provides reference configurations validated through internal performance testing.

Scale TierManaged ClustersNumber of NodesCPU per NodeMemory per NodeNotes
Small≤ 1038 cores16 GBSuitable for small-scale environments
Medium≤ 50316 cores32 GBDefault production setup
Large≤ 100324 cores48 GBSupports heavy Web Console usage and frequent sync cycles
Extra Large≤ 500632 cores64 GBRequires horizontal scaling and dedicated infra nodes
WARNING

These recommendations are general guidelines. Actual requirements depend on your cluster topology, user concurrency, and plugins installed.

Vertical Scaling Guidelines

When increasing load per node (for example, 2× more clusters or higher user concurrency), follow these adjustments:

ParameterScaling Recommendation
CPU+50% for every 50 additional managed clusters
Memory+50% for every 50 additional managed clusters

Horizontal Scaling Guidelines

When exceeding 100 managed clusters or encountering persistent API latency above 500 ms:

Add nodes to distribute request handling and controller workloads.

Resource Validation and Monitoring

After deployment, continuously monitor the following metrics to validate node sizing:

MetricRecommended Range
Node CPU utilization60–75% under peak load
Node Memory utilization≤80% sustained
API request latencyP90 < 500ms
etcd commit latencyP99 < 50ms
Node CPU utilization
Node Memory utilization
API request latency
etcd commit latency
100 * (1 - avg by (instance)(rate(node_cpu_seconds_total{mode="idle"}[5m])))
NOTE

If sustained resource usage consistently exceeds recommended thresholds, scale vertically (add CPU/memory) or horizontally (add nodes) before user-facing performance degradation occurs.

Global Cluster as a Workload Cluster

In a Single Cluster deployment, the global cluster also serves as the workload cluster. Size its control plane nodes by the number of worker nodes with the following table. The workload cluster table in Evaluating Resources for Workload Cluster does not apply directly, for these reasons:

  • The global cluster runs the platform management components in addition to the control plane of the workload cluster.
  • As the number of nodes grows, the load on the global cluster grows faster than the node count. Larger clusters usually run more features, hold more custom resources, and send more requests to the API server.
Number of worker nodesCluster-density (namespaces)CPU coresMemory (GB)
24500816
12010001632
254400032128

The headroom guidance for workload clusters also applies: target an overall CPU and memory utilization of about 60% or less; see Control Plane Node Sizing.

Summary

When sizing the global cluster of a Multi-Cluster deployment:

  1. Begin with 3 nodes × 16 cores × 32 GB for moderate-scale deployments (≤50 clusters).
  2. Scale vertically for higher request concurrency or heavy Web Console usage.
  3. Scale horizontally beyond 100 clusters to maintain API responsiveness.
  4. Re-evaluate sizing after every significant increase in managed cluster count or sync frequency.

Use these practices to keep resource planning aligned with the growth of your Multi-Cluster environment.