Clarification on CPU % in VARZ for NATS running on AWS Fargate

[NevinDry]

Hello NATS community,

I’m working on a healthcheck script for NATS using the CLI command:

nats server check server

I want to set CPU warning and critical thresholds based on the number of CPUs. Previously, I was using nproc to detect the number of cores and calculate thresholds. For example, in a non-containerized environment, if I have 2 cores and want a warning threshold at 80%, I might set it as 2 * 80 = 160% (assuming the monitoring tool sums CPU usage across all cores).

However, I noticed an interesting behavior when running NATS on AWS ECS Fargate. Consider a container with 0.25 vCPU. When querying the VARZ endpoint, I get something like:

{
...
"cores": 2,
"cpu": 10,
...
}

The "cores": 2 value corresponds to the number of logical CPUs on the host, not the vCPU allocated to my container. This is consistent with nproc, which also shows 2.

My main question:

Is the "cpu" value from VARZ (10 in this example) calculated relative to the allocated vCPU (0.25) or to the logical cores on the host (2)?

This distinction is critical because it changes how I should calculate CPU thresholds for alerts. If VARZ CPU % reflects actual usage relative to the container’s vCPU allocation, I should base thresholds on the container CPU configuration rather than the nproc / cores value.

Has anyone experienced similar monitoring or alerting issues on AWS Fargate because of this? Any guidance on best practices for CPU-based health checks in containerized NATS would be greatly appreciated.

Thank you very much!