kubectl top pods filter by the node

[taraspos]

What would you like to be added:

I would like to be able to see resource consumption for all pods running on the specific node only.

kubectl top pods --field-selector spec.nodeName=<name>

Currently this command is failing with:

➜  ~ kubectl top pods --field-selector spec.nodeName=<node> -A
Error from server (BadRequest): "spec.nodeName" is not a known field selector: only "metadata.name", "metadata.namespace"

Why is this needed:

This is useful when investigating node resource pressure. Let's say that I run kubectl top nodes --sort-by='memory' and see that one of my nodes is on 95% memory usage. The next logical thing to do is to find what are the most memory intensive pods.

I can run kubectl describe node <name>, this will give me list of pods and their memory requests and limits but not the usage.

Currently this can be fetched by the following overly-complicated command:

kubectl get pods --field-selector spec.nodeName='<name>' -A -o custom-columns=NAMESPACE:.metadata.namespace,NAME:.metadata.name --no-headers | \ 
   xargs -I {} sh -c "kubectl top pod --no-headers -n {}" | \
   column -t

Related issues:

• add NODE info for kubectl top pod output kubectl#1706
• Add field-selector option for kubectl top pod kubectl#923

[k8s-ci-robot]

This issue is currently awaiting triage.

SIG CLI takes a lead on issue triage for this repo, but any Kubernetes member can accept issues by applying the triage/accepted label.

The triage/accepted label can be added by org members by writing /triage accepted in a comment.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

[googs1025]

/assign
This seems reasonable, I will look at the feasibility of adding a flag

[mpuckett159]

@ardaguclu is going to reach out. to sig-instrumentation because they are in charge of what data is available in the Metrics API (the API that is communicated with to get the information displayed in kubectl top pods) and if they are willing to add this field to the pod metrics data we can move forward with this.

If not, we will close this issue and you can watch kubernetes/kubectl#1706 to be closed and that will simplify your bash command by allowing you to grep the output easily.

[ardaguclu]

/transfer kubernetes

[ardaguclu]

/sig instrumentation
/sig cli

[ardaguclu]

@kubernetes/sig-instrumentation-leads we need your input in here. There is an attempt #130525 that manually fetches the node of the pod name and links the node name to the output of retrieved from metrics API. Is this correct?. Is there any similar discussions about adding a node name field in metrics API for "top pods"?.