View metrics
The centron Metrics Agent is pre-installed on all worker nodes. You do not need to install anything to analyse your cluster’s utilisation – the data is available directly in Control Panel.
Prerequisites
- Access to the centron Control Panel
- A Kubernetes cluster with the status Running and a verified connection via
kubectl
Viewing metrics in Control Panel
- Open the cluster in the Kubernetes overview.
- Switch to the Analytics tab.
Four diagrams are displayed:
| Diagram | Meaning |
|---|---|
| CPU Usage | Processor utilisation of the nodes, broken down into sys (system processes) and user (your workloads). Persistently high values indicate that the nodes are undersized. |
| Memory | Memory utilisation. If it reaches the capacity limit, pods are evicted (OOMKilled). |
| Disk I/O | Read and write accesses to the local NVMe storage. |
| Bandwidth | Network throughput of the nodes. |
Narrow down the display
There are two selection fields above the charts:
- Node pool: By default, All node pools. Select a single pool to view its utilisation in isolation.
- Time period: You can choose between 1 hour, 6 hours and 24 hours.
Comparing individual pools is useful when a cluster contains several pools of different sizes: this allows you to determine whether a peak in workload affects the entire cluster or only a specific workload.
Querying metrics via kubectl
In addition, you can retrieve the current utilisation directly via kubectl. Unlike the Analytics tab, which shows trends over time, these commands provide snapshots:
# Auslastung je Node
kubectl top nodes
# Auslastung je Pod über alle Namespaces
kubectl top pods --all-namespaces
# Pods mit der höchsten CPU-Last zuerst
kubectl top pods --all-namespaces --sort-by=cpu
To get an idea of how many resources are already reserved on a node:
kubectl describe node <node-name>
The Allocated resources section shows the total amount of requests requested – this determines whether further pods can be placed on the node, regardless of the actual utilisation.
Putting values into perspective
- High CPU load, low RAM: A cluster’s CPU utilisation naturally fluctuates significantly. Short-lived peaks are not a cause for concern; sustained utilisation close to the upper limit suggests a need for larger nodes or more nodes in the pool.
- Memory at capacity: Memory cannot be compressed. If a node reaches its limit, Kubernetes terminates pods. Ensure you maintain a deliberate reserve here.
- High utilisation despite spare capacity: Check the
requestssettings. Values set too high block capacity that is not actually being used.