Overview
When monitoring Kubernetes workloads using AppDynamics' Cluster Agent, users often observe a mismatch or confusion between the Memory (MB) values in the table and the Usage% shown in pod detail graphs. This article explains how CPU and memory metrics are computed, displayed, and what to expect when resource limits are or are not defined.
Ask 1: How is CPU% Busy calculated for Pods?
AppDynamics uses the following logic:
-
If CPU limits are defined in the pod spec, the CPU%Busy is calculated relative to the defined limit.
-
If limits are not set, AppDynamics falls back to node-level CPU usage, and percentage values may not reflect actual pod utilization.
Ask 2: Does the same logic apply to Memory (MB)?
Not exactly.
-
Memory (MB) is always collected as the raw value used by the container.
-
Memory usage % (Usage%) is only calculated and displayed if a memory limit is defined.
-
If no memory limit is set, AppDynamics may still show MB used in the table view, but the Usage% will remain at 0%, as no denominator is available for calculation.
Ask 3: Why does the graph show 0% even when memory is consumed?
This behavior is expected when no resource limits are set. AppDynamics cannot calculate a percentage of memory used if there's no defined upper limit to compare against.
Screenshot Summary:
-
Screenshot 1: Pod list view
-
Shows the AppDynamics Operator pod using 11.84 MB memory with 1% CPU.
-
2. Screenshot 2: Pod container view (with limits defined)
-
Graph shows 11.91% memory usage out of 100Mi defined.
3. Screenshot 3: Pod container view (without limits)
-
NodeJS pod consuming 33 MB memory but graph shows 0% usage.
Verifying Pod Resource Limits:
To check if a pod has limits defined:
kubectl get pod <pod-name> -n <namespace> -o jsonpath="{.spec.containers[*].resources.limits}"
If output is empty ({}), no limits are set, and usage% will show 0%.
Conclusion:
-
Always refer to the table for actual memory used (MB).
-
Usage% depends on resource limits being set.
-
No bug here – this is by design, and understanding this helps interpret the UI accurately.
Recommendations:
-
Set CPU and memory limits in pod specs for better visibility and scaling insights.
-
Use kubectl top pod to cross-verify values if needed.
