# Bookstore — Part 12 ch.03 "Batch and gang scheduling": a Kueue
# ResourceFlavor = a NAMED KIND OF CAPACITY the ClusterQueue draws quota from.
#
# !!! CRD-INTRINSIC DRY-RUN (identical precedent to raw-manifests/51-/70-/83-,
#     argocd/, operators/, chaos/) !!!
#   `ResourceFlavor` is a Kueue CRD (kueue.x-k8s.io/v1beta2). WITHOUT Kueue
#   installed a client dry-run prints:
#     no matches for kind "ResourceFlavor" in version "kueue.x-k8s.io/v1beta2"
#   EXPECTED and SCHEMA-CORRECT — install Kueue first (Part 12 ch.03 Hands-on:
#   pinned Helm `oci://registry.k8s.io/kueue/charts/kueue`, own namespace
#   `kueue-system`). A whole-dir dry-run prints this for CRD-backed files only
#   and continues. Schema verified against Kueue kueue.x-k8s.io/v1beta2
#   ResourceFlavor.
#
# This DEFAULT flavor has no nodeLabels, so it matches ANY node — that is what
# lets the CPU-only recommendations "training" demo run on kind with no GPU.
#
# FOR A REAL GPU POOL you would add a SECOND flavor pinned to GPU nodes, e.g.:
#   apiVersion: kueue.x-k8s.io/v1beta2
#   kind: ResourceFlavor
#   metadata: { name: bookstore-ml-gpu-flavor }
#   spec:
#     nodeLabels:
#       nvidia.com/gpu.present: "true"   # GPU Operator / GFD label (ch.02)
#   ...and reference it from the ClusterQueue's nvidia.com/gpu resourceGroup so
#   GPU quota is accounted against GPU nodes specifically.
apiVersion: kueue.x-k8s.io/v1beta2
kind: ResourceFlavor
metadata:
  name: bookstore-ml-flavor
  labels:
    app.kubernetes.io/part-of: bookstore-ml
# No spec.nodeLabels: matches any node (CPU-runnable on kind). The GPU-pinned
# flavor variant is documented in the header comment above.