Virtual memory pointing value of namespace rather than pod

[human-77]

Description

I was creating some dashboards to monitor an instance of frontend running on a pod when I found this bizarre value for the virtual memory

# HELP frontend_process_virtual_memory_bytes Virtual memory size in bytes.
# TYPE frontend_process_virtual_memory_bytes gauge
frontend_process_virtual_memory_bytes 33012293632

Using kubectl describe ns my-namespace I got:

limits.memory                     9829Mi  30Gi
requests.cpu                      1800m   10
requests.ephemeral-storage        0       10Gi
requests.memory                   2560Mi  30Gi
requests.storage                  0       30Gi

In other words, despite running on a pod, the virtual memory is pointing the virtual memory of the WHOLE NAMESPACE

Link to the page

private site

Steps to reproduce

We are running it on a cloud using helm charts. I believe that you can simply run the pod on a namespace meant for multiple pods and the error should happen.

App version

v1.36.0

Browser

Doesn't matter in this case.

Operating system

Ubuntu docker image

Additional information

No response

[tom2drum]

Hi @human-77. We use the standard prom-client library to obtain this metric. I believe a bug lies on their side, specifically in the area where the metrics are collected. In our app, we simply call the collectDefaultMetrics method from the library.

[human-77]

So what should I do?

Do I open a issue there or there is already something about the issue?

[tom2drum]

Yes, I suggest to open a discussion/issue in the prom-client repo.

[human-77]

Thank for the help!

[human-77]

For anyone with the same problem, I opened a new issue HERE.