[Bug]: Is worker_processes of NGINX Ingress automatically set according to node’s physical cores instead of pod CPU limit?

[SonNgo2211]

Version

5.0.0

What Kubernetes platforms are you running on?

Other

Steps to reproduce

Hi team,

I read the discussion at kubernetes/ingress-nginx#11518 regarding the worker_processes setting in ingress-nginx. It seems that when worker_processes is set to auto, the number of worker processes is determined by the number of physical CPU cores of the node, not by the CPU limit assigned to the NGINX pod/container in Kubernetes. This leads to a situation where a pod with e.g. 1.5 CPU limit may spawn many workers if the node is large, possibly causing CPU throttle and inefficiency.

• Does NGINX (inside the controller pod) automatically detect host physical cores, or does it respect the container's CPU limit when calculating worker processes in auto mode?
• How does the controller recommend setting worker_processes for best efficiency and compatibility with CPU-limited pods?

Any suggestions or insights from the community would be highly appreciated.
Thank you all in advance!

[github-actions]

Hi @SonNgo2211 thanks for reporting!

Be sure to check out the docs and the Contributing Guidelines while you wait for a human to take a look at this 🙂

Cheers!

[haywoodsh]

The discussion you linked suggests that if worker_processes is set to auto, the community ingress would fetch or compute an actual number when creating the NGINX config.
However, there logic does not exist in the NGINX Ingress Controller. Therefore, if the config map value of worker_processes is auto, the NGINX Ingress Controller will simply write the string auto in the config, and it will be up to NGINX itself will be responsible for detecting the number of CPU cores.
Based on this Github comment and the fact that extra logic was required in ingress-nginx to adjust the value, NGINX might not read cgroups files or the container CPU limit. I'll still have to confirm this, but if you want to align the value of worker_processes with the pod's CPU limit, you can try setting the config map key worker_processes to a specific number.

[SonNgo2211]

I'll still have to confirm this, but if you want to align the value of worker_processes with the pod's CPU limit, you can try setting the config map key worker_processes to a specific number.

Yes I did that. I think nginx ingress also need logic to set limit for worker process according to cpu limit pod. This may lead to resource consumption and may cause errors that we don't expect.

[haywoodsh]

As discussed, the worker_processes directive is a core NGINX setting, and the Ingress Controller is designed to pass this configuration through directly. Adding a custom translation layer here would create a non-standard behaviour that is not possible for us to maintain across different Kubernetes environments.

The best path forward could be improving how NGINX itself handles auto in containers. I recommend proposing this directly to the NGINX core team. Once implemented upstream, we will inherit the capability in our next NGINX version bump.

[SonNgo2211]

Thanks @haywoodsh, I'll probably create an issue at the NGINX core project, hardsetting worker_processes is a good solution for now.