Create extended resources with NodeFeatureRule

[marquiz]

What would you like to be added:

In addition to labels, taints (#540) and annotations (#863), support also management of extended resources with the NodeFeatureRule CRD API.

Why is this needed:

There are usage scenarios where people want to advertise features as extended resources instead of labels (or annotations), e.g. #1079. There has also been use cases like accounting number of network bridge interfaces (or ports, I don't remember) and GPU memory for example. We have long had experimental support for extended resources via the -resource-labels command line flag in nfd-master but that is very unwieldy for users (but obviously has worked as we haven't basically got any bug reports or complaints about that).

[marquiz]

ping @uniemimu @mythi @fidencio

[marquiz]

/help

On top of my head an outline/list of sub-tasks related to the implementation:

1. Extend the NodeFeatureRule CRD API and helpers
   1. Add new extendedResources field to NodeFeatureRuleSpec in pkg/apis/nfd/v1alpha1/types.go
   2. Run make generate to update yamls and auto-generated code
   3. simple update of the RuleOutput type (and related methods/funcs) in pkg/apis/nfd/v1alpha1/rule.go to return also extended resources (in addition to labels and taints)
2. Implement support in nfd-master (not super complex as the basic infrastructure for handling extended resources is already in place)
   1. update processNodeFeatureRule() to return extended resources
   2. update refreshNodeFeatures() to correspondingly update the node status with these extended resources
3. Documentation: at least introduction.md and customization-guide.md
4. Extend e2e-tests to cover extended resources

Few extra considerations regarding "namespacing" i.e. prefixing of names of extended resources:

1. We probably want to specify a "default namespace" that gets applied if no namespace (i.e. <namespace>/ prefix is give). Basically add a new const like ExtendedResourceNs in pkg/apis/nfd/v1alpha1/annotations_labels.go. Could be e.g. the same as with labels i.e. feature.node.kubernetes.io or nfd.kubernetes.io (dunno what would make most sense).
2. It might be best to just allow any vendor specific namespaces, but might want to deny kubernetes.io/ and its sub-namespaces (*.kubernetes.io/) (not sure about this either)

EDIT: fixed misspelling NodeFeatureSpec -> NodeFeatureRuleSpec

[k8s-ci-robot]

@marquiz:
This request has been marked as needing help from a contributor.

Guidelines

Please ensure that the issue body includes answers to the following questions:

• Why are we solving this issue?
• To address this issue, are there any code changes? If there are code changes, what needs to be done in the code and what places can the assignee treat as reference points?
• Does this issue have zero to low barrier of entry?
• How can the assignee reach out to you for help?

For more details on the requirements of such an issue, please see here and ensure that they are met.

If this request no longer meets these requirements, the label can be removed
by commenting with the /remove-help command.

In response to this:

/help

On top of my head an outline/list of sub-tasks related to the implementation:

1. Extend the NodeFeatureRule CRD API and helpers
   1. Add new extendedResources field to NodeFeatureSpec in pkg/apis/nfd/v1alpha1/types.go
   2. Run make generate to update yamls and auto-generated code
   3. simple update of the RuleOutput type (and related methods/funcs) in pkg/apis/nfd/v1alpha1/rule.go to return also extended resources (in addition to labels and taints)
2. Implement support in nfd-master (not super complex as the basic infrastructure for handling extended resources is already in place)
   1. update processNodeFeatureRule() to return extended resources
   2. update refreshNodeFeatures() to correspondingly update the node status with these extended resources
3. Documentation: at least introduction.md and customization-guide.md
4. Extend e2e-tests to cover extended resources

Few extra considerations regarding "namespacing" i.e. prefixing of names of extended resources:

1. We probably want to specify a "default namespace" that gets applied if no namespace (i.e. <namespace>/ prefix is give). Basically add a new const like ExtendedResourceNs in pkg/apis/nfd/v1alpha1/annotations_labels.go. Could be e.g. the same as with labels i.e. feature.node.kubernetes.io or nfd.kubernetes.io (dunno what would make most sense).
2. It might be best to just allow any vendor specific namespaces, but might want to deny kubernetes.io/ and its sub-namespaces (*.kubernetes.io/) (not sure about this either)

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/test-infra repository.

[ArangoGutierrez]

/assign

[PiotrProkop]

@marquiz @ArangoGutierrez Another example of how it can be useful(in other words we would definitely use it!) is to expose Intel RDT num_closid the max amount of RDT control groups in the system (prolly -1 for default group).