DRA: Derived Attributes by gauravkghildiyal · Pull Request #6081 · kubernetes/enhancements

gauravkghildiyal · 2026-05-15T00:23:13Z

One-line PR description: Add KEP-6080 introducing derivedAttributes for flexible DRA device co-allocation.

Issue link: DRA: Derived Attributes #6080

Other comments: This KEP introduces scoped CEL expressions (derivedAttributes) to DeviceRequest to synthesize virtual grouping keys on the fly.

/sig scheduling
/wg device-management

k8s-ci-robot · 2026-05-15T00:23:19Z

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: gauravkghildiyal
Once this PR has been reviewed and has the lgtm label, please assign sanposhiho for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

keps/sig-scheduling/OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

johnbelamaric · 2026-05-15T00:43:25Z

+      - kube-apiserver
+      - kube-controller-manager
+      - kube-scheduler
+      - kubelet


I doubt kubelet needs to know about these

Makes sense removed.

(I'm not too sure about kube-controller-manager either, but maybe its needed there because it clones the ResourceClaimTemplate to ResourceClaims)

johnbelamaric · 2026-05-15T00:44:55Z

+## Proposal
+
+We propose extending `.devices.requests` with `derivedAttributes` and
+`.devices.constraints` with `matchDerivedAttribute`. A derived attribute defines


We should not change the constraints. These should just look like any other attribute to the entire constraints section. The only API addition needed should be the derivedAttributes field.

Great idea! Even better.

Updated.

johnbelamaric · 2026-05-15T00:45:54Z

@thameem-abbas I think this might be useful for your use cases.

johnbelamaric · 2026-05-15T00:46:19Z

cc @pohly

…ttribute

pravk03 · 2026-05-15T03:21:12Z

/cc

pohly

Looks okay to me.

Shadow API review: we need to sort out some details, see comments.

pohly · 2026-05-15T12:47:21Z

+- **Risk**: A new CEL expression needs to be evaluated for each candidate device
+  (in addition to any CEL expressions evaluated for device selectors).
+- **Mitigation**: This evaluation adds a constant time increment to each device
+  evaluation. The remainder of the scheduling cycle—including constraint


It's constant per device, but each device might be considered more than once. Caching derived attributes mitigates that.

pohly · 2026-05-15T12:48:01Z

+type DerivedAttribute struct {
+	// Name is the identifier for this derived attribute, used in constraints.
+	//
+	// +required


Document the valid string format here.

pohly · 2026-05-15T12:48:14Z

+	// for each candidate device.
+	// +featureGate=DRADerivedAttributes
+	// +optional
+	DerivedAttributes []DerivedAttribute `json:"derivedAttributes,omitempty"`


Size limit?

pohly · 2026-05-15T12:51:42Z

+
+	// Expression is a CEL expression evaluated against each candidate device.
+	// The expression must evaluate to a primitive scalar (string, integer,
+	// boolean) or a list of scalars ([]string, []int64, []bool) to act as a


We also have semver. Are those valid results?

pohly · 2026-05-15T12:54:37Z

+- **Environment**: The CEL environment for `Expression` is exactly the same
+  as that for `CELDeviceSelector`, containing a single variable `device`.
+- **Return Type**: The CEL expression must evaluate to a scalar (string,
+  integer, boolean) or a list of scalars (`[]string`, `[]int64`, `[]bool`).


This "must evaluate to" cannot be validated during admission time (depends on unknown input types).

How are evaluation errors at runtime handled?

pohly · 2026-05-15T12:56:21Z

+	//    of the attributes which were prefixed by "dra.example.com").
+	//  - capacity (map[string]object): the device's capacities, grouped by prefix.
+	//
+	// +required


Size limit?

aojea · 2026-05-15T14:59:11Z

+  - "@gauravkghildiyal"
+owning-sig: sig-scheduling
+participating-sigs:
+  - sig-network


I can review from sig network

aojea · 2026-05-15T15:00:36Z

+object. The resulting value acts as a virtual attribute that can be referenced
+directly by existing `.devices.constraints[].matchAttribute` fields.
+
+### Core Manifest Example: GPU & NIC NUMA Alignment


This is a great example , cc: @ffromani @pravk03

aojea · 2026-05-15T15:07:55Z

+   syntactic boundary between static (FQDN) and derived (bare) attributes,
+   eliminating shadowing risks but preventing direct attribute overrides.
+
+Recommended: **Option 1 (Allowing FQDNs)**. Most manifest authors will naturally


I also think that Option 2 will break existing users of the well defined pcieRoot standard attribute, if a new driver that does not implement the standard attribute want to use it I think it should have to override it with the FQDN

yongruilin

Declarative Validation plans to support validation for new API fields without handwritten validation counterpart directly.
But for complicated validation(e.g. CEL compilable), might still need to reply on handwritten.
/cc @jpbetz

yongruilin · 2026-05-15T21:21:57Z

+	// +optional
+	DerivedAttributes []DerivedAttribute `json:"derivedAttributes,omitempty"`


Suggested change

// +optional

DerivedAttributes []DerivedAttribute `json:"derivedAttributes,omitempty"`

// +optional

// +k8s:optional

// +k8s:maxItems=<int>

DerivedAttributes []DerivedAttribute `json:"derivedAttributes,omitempty"`

kad · 2026-05-18T14:07:09Z

/cc

k8s-ci-robot added sig/scheduling Categorizes an issue or PR as relevant to SIG Scheduling. wg/device-management Categorizes an issue or PR as relevant to WG Device Management. labels May 15, 2026

github-project-automation Bot added this to Dynamic Resource Allocation and SIG Scheduling May 15, 2026

github-project-automation Bot moved this to Needs Triage in SIG Scheduling May 15, 2026

github-project-automation Bot moved this to 🆕 New in Dynamic Resource Allocation May 15, 2026

k8s-ci-robot added the kind/kep Categorizes KEP tracking issues and PRs modifying the KEP directory label May 15, 2026

k8s-ci-robot requested review from dom4ha and macsko May 15, 2026 00:23

k8s-ci-robot added cncf-cla: yes Indicates the PR's author has signed the CNCF CLA. size/XL Denotes a PR that changes 500-999 lines, ignoring generated files. labels May 15, 2026

gauravkghildiyal mentioned this pull request May 15, 2026

DRA: Derived Attributes #6080

Open

4 tasks

gauravkghildiyal force-pushed the dra-derived-attributes branch from 217ed13 to 7d68c53 Compare May 15, 2026 00:33

Add DRA Derived Attributes KEP

df713b1

gauravkghildiyal force-pushed the dra-derived-attributes branch from 7d68c53 to df713b1 Compare May 15, 2026 00:35

johnbelamaric reviewed May 15, 2026

View reviewed changes

Address John's comment about not introducing a separate matchDerivedA…

9d3485f

…ttribute

k8s-ci-robot requested a review from pravk03 May 15, 2026 03:21

pohly reviewed May 15, 2026

View reviewed changes

aojea reviewed May 15, 2026

View reviewed changes

yongruilin reviewed May 15, 2026

View reviewed changes

k8s-ci-robot requested a review from jpbetz May 15, 2026 21:27

pohly moved this from 🆕 New to 👀 In review in Dynamic Resource Allocation May 18, 2026

k8s-ci-robot requested a review from kad May 18, 2026 14:07

liggitt added this to @liggitt May 20, 2026

liggitt moved this to Todo in @liggitt May 20, 2026

liggitt added this to API Reviews May 20, 2026

liggitt moved this to Assigned in API Reviews May 20, 2026

		// +optional
		DerivedAttributes []DerivedAttribute `json:"derivedAttributes,omitempty"`

Conversation

gauravkghildiyal commented May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

k8s-ci-robot commented May 15, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

johnbelamaric commented May 15, 2026

Uh oh!

johnbelamaric commented May 15, 2026

Uh oh!

pravk03 commented May 15, 2026

Uh oh!

pohly left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yongruilin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kad commented May 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

gauravkghildiyal commented May 15, 2026 •

edited

Loading