Add K8s pod options: hostNetwork, privileged, hostIPC, shmSize, extraResources (#85, #86, #87, #88)

[powderluv]

Summary

Fixes #85, #86, #87, #88 — adds five K8s pod options required for GPU training workloads.

Changes

Thread five new fields through the CRD → core JobSpec → proto → pod creation pipeline:

Field	Issue	Pod Effect
hostNetwork	#85	PodSpec.host_network = true
privileged	#86	Container.security_context.privileged = true
hostIpc	#87	PodSpec.host_ipc = true
shmSize	#87	emptyDir volume at /dev/shm with sizeLimit
extraResources	#88	Added to pod resource requests/limits (e.g., rdma/devices: 1)

Example SpurJob

apiVersion: spur.ai/v1alpha1
kind: SpurJob
spec:
  image: nvcr.io/nvidia/pytorch:latest
  hostNetwork: true
  privileged: true
  hostIpc: true
  shmSize: "64Gi"
  extraResources:
    rdma/hca_shared_devices_a: "1"
  gpus:
    count: 8
    gpuType: mi300x

Test plan

• Full test suite passes (0 failures)
• Existing CRD parse tests updated with new fields
• K8s integration: deploy SpurJob with hostNetwork=true, verify pod has host networking

@amd-vpenumal — these are the fields you requested. Please verify the CRD schema matches your expected usage.

🤖 Generated with Claude Code

[shiv-tyagi]

Just one thought. We can merge though.

[shiv-tyagi]

Note to self:

Does the extra device plugin resources field would only make sense for k8s or spur in general? If later is not true, it might we worth keeping them within spur-k8s somehow.

I will take a look on this later.