kubeflow / trainer / 38453510684

77%

Build Type

push

github

Committed by web-flow

Commit Message

feat(api): implement DRA support for TrainJob (KEP-2782) (#4027)

* feat(api): add resourceClaimsPerNode, runtimePatches DRA fields and DynamicResourceAllocation gate

Adds Trainer.resourceClaimsPerNode (template-only, DNS-label name), PodSpecPatch.resourceClaims and ContainerPatch.resources per KEP-2782, rejects resourcesPerNode.claims via CEL, and introduces the DynamicResourceAllocation alpha feature gate.

Signed-off-by: Sridhar1030 <sridharpillai75@gmail.com>

* test(integration): cover DRA claims wiring and webhook validation

Enables the DynamicResourceAllocation gate in the integration framework and adds controller and webhook specs for resourceClaimsPerNode, runtimePatches claims and dangling-claim rejection. The controller specs pass once the runtime wiring lands.

Signed-off-by: Sridhar1030 <sridharpillai75@gmail.com>

* feat(runtime): wire resourceClaimsPerNode into the trainer node PodSpec

Applies the TrainJob claims to the node Pod's resourceClaims and the node container's resources.claims, keeping TrainJob claims ahead of runtime-wired ones and preserving an existing container claim's request. Adds the operator guide.

Signed-off-by: Sridhar1030 <sridharpillai75@gmail.com>

* feat(runtime): support DRA claims through runtimePatches

Documents PodSpecPatch.resourceClaims and ContainerPatch.resources for sidecars and init containers and covers the combined resourceClaimsPerNode plus runtimePatches path. Strategic merge already handles the new fields.

Signed-off-by: Sridhar1030 <sridharpillai75@gmail.com>

* feat(webhooks): validate DRA fields behind the feature gate

Rejects DRA fields on create and on updates that introduce them while the gate is disabled, rejects runtime templates and runtimePatches whose container claims reference no Pod-level claim.

Signed-off-by: Sridhar1030 <sridharpillai75@gmail.com>

* fix(runtime): address review feedback on DRA claims wiring

Signed-off-by: Sridhar1030 <sridharpillai75@gmail.com>

* test(inte... (continued)

Coverage Stats

169 of 199 new or added lines in 6 files covered. (84.92%)

1 existing line in 1 file now uncovered.

3511 of 4791 relevant lines covered (73.28%)

0.85 hits per line