#!/usr/bin/env bash
# nvidia/gpu-operator v26.7.1 - base variant: preinstalled-driver-and-toolkit
# Path: fetch this entry's retained objects, check their SHA-256, then apply
# them with kubectl. It uses curl, shasum, awk and kubectl. It needs no cub
# CLI and no ConfigHub account.
# Generated by scripts/generate-public-site.mjs from this entry's listing:
# https://confighub.github.io/helm-expt/site/listings/nvidia-gpu-operator-v26-7-1-preinstalled-driver-and-toolkit.json
# Entry page: https://confighub.github.io/helm-expt/site/charts/nvidia-gpu-operator-v26-7-1.html
set -euo pipefail

say() { printf '\n>> %s\n' "$*"; }

for tool in curl awk kubectl; do
  if ! command -v "$tool" >/dev/null 2>&1; then
    printf '%s is required and was not found.\n' "$tool" >&2
    exit 1
  fi
done
if command -v shasum >/dev/null 2>&1; then
  sha256_of() { shasum -a 256 "$1" | awk '{print $1}'; }
elif command -v sha256sum >/dev/null 2>&1; then
  sha256_of() { sha256sum "$1" | awk '{print $1}'; }
else
  printf 'shasum is required and was not found.\n' >&2
  exit 1
fi

work=./nvidia-gpu-operator-v26-7-1-preinstalled-driver-and-toolkit-objects
objects_url=https://raw.githubusercontent.com/confighub/helm-expt/main/recipes/nvidia/gpu-operator/v26.7.1/revisions/preinstalled-driver-and-toolkit/r001/rendered/release-objects.yaml
objects_sha256=2b6cb805960b912ddebd8e5ead4b9bfe5a75ec8d6e28816ebcc9173bdf5da82b

say "Fetch the retained objects into $work/objects.yaml"
mkdir -p "$work"
curl -fsSL -o "$work/objects.yaml" "$objects_url"

say "Check the file against the SHA-256 in the listing"
fetched_sha256="$(sha256_of "$work/objects.yaml")"
if [ "$fetched_sha256" != "$objects_sha256" ]; then
  printf 'The file differs from the listing, so the script stops. Nothing was applied.\n  listing: %s\n  fetched: %s\n' "$objects_sha256" "$fetched_sha256" >&2
  exit 1
fi
printf 'The file matches: sha256 %s\n' "$fetched_sha256"

say "Split the file into its 27 objects. The cluster is untouched so far"
rm -rf "$work/docs" "$work/crds" "$work/rest"
mkdir -p "$work/docs" "$work/crds" "$work/rest"
awk -v dir="$work/docs" 'BEGIN { n = 0; f = sprintf("%s/doc-%04d.yaml", dir, n) } /^---[ \t]*$/ { close(f); n++; f = sprintf("%s/doc-%04d.yaml", dir, n); next } { print > f }' "$work/objects.yaml"
for n in 0002 0004 0006 0008 0010 0012 0013 0014; do mv "$work/docs/doc-$n.yaml" "$work/crds/"; done
for n in 0015 0016 0017 0018 0019 0020 0021 0022 0023 0024 0025 0026; do mv "$work/docs/doc-$n.yaml" "$work/rest/"; done
for n in 0027 0028 0029 0030 0031 0032 0033; do mv "$work/docs/doc-$n.yaml" "$work/rest/"; done
ls "$work/crds" "$work/rest"

say "Check that kubectl reaches a cluster"
if ! kubectl get --raw /version >/dev/null 2>&1; then
  printf 'kubectl cannot reach a cluster. Point it at a throwaway cluster and re-run this script.\n' >&2
  exit 1
fi
kubectl config current-context 2>/dev/null || true

say "Create the gpu-operator namespace if it does not exist"
if ! kubectl get namespace gpu-operator >/dev/null 2>&1; then
  kubectl create namespace gpu-operator
fi

wait_for_crd() {
  crd_name="$1"
  deadline=$(( $(date +%s) + 180 ))
  until kubectl get "crd/${crd_name}" >/dev/null 2>&1; do
    if [ "$(date +%s)" -ge "$deadline" ]; then
      printf "CRD %s did not appear within 180 seconds.\n" "$crd_name" >&2
      return 1
    fi
    sleep 2
  done
  kubectl wait --for=condition=Established --timeout=120s "crd/${crd_name}"
}

say "Check the 8 CRDs this file holds"
missing_crds=0
if ! kubectl get crd/clusterpolicies.nvidia.com >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/gpuclusters.nvidia.com >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/nvidiadrivers.nvidia.com >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/computedomaincliques.resource.nvidia.com >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/computedomains.resource.nvidia.com >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/nodefeatures.nfd.k8s-sigs.io >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/nodefeaturegroups.nfd.k8s-sigs.io >/dev/null 2>&1; then
  missing_crds=1
fi
if ! kubectl get crd/nodefeaturerules.nfd.k8s-sigs.io >/dev/null 2>&1; then
  missing_crds=1
fi
if [ "$missing_crds" -eq 1 ]; then
  say "A CRD is missing, so apply the file's CRDs first"
  kubectl apply --server-side -f "$work/crds"
  crd_outcome="were applied first"
elif kubectl diff --server-side -f "$work/crds" >/dev/null 2>&1; then
  say "The CRDs already exist and match this file. The script leaves them under their current owner"
  crd_outcome="already existed and matched, so the script left them alone"
elif [ "${APPLY_CRDS:-0}" = "1" ]; then
  say "The existing CRDs differ from this file. APPLY_CRDS=1 is set, so the script applies the file's CRDs"
  kubectl apply --server-side -f "$work/crds"
  crd_outcome="differed and were applied, because APPLY_CRDS=1 was set"
else
  say "The CRDs already exist and differ from this file, or another tool owns them"
  cat <<'EOF_CRDS'
This script does not change CRDs it did not just create.
If this is an upgrade, the objects that follow may need the newer CRDs.
To let this script apply the file's CRDs, re-run with:
  APPLY_CRDS=1 bash try-objects.sh
EOF_CRDS
  exit 1
fi
wait_for_crd clusterpolicies.nvidia.com
wait_for_crd gpuclusters.nvidia.com
wait_for_crd nvidiadrivers.nvidia.com
wait_for_crd computedomaincliques.resource.nvidia.com
wait_for_crd computedomains.resource.nvidia.com
wait_for_crd nodefeatures.nfd.k8s-sigs.io
wait_for_crd nodefeaturegroups.nfd.k8s-sigs.io
wait_for_crd nodefeaturerules.nfd.k8s-sigs.io

say "Apply the other 19 objects"
kubectl apply -f "$work/rest"

say "Done. The 8 CRDs $crd_outcome. kubectl applied the other 19 objects of this entry's retained file."
say "The script stops here. It did not wait for any workload to start"
say "Not run by this script: 3 steps of this entry's plan. The entry page says who runs each one."
cat <<'EOF_NOT_RUN'
   - Before an upgrade, move the CRDs to the new version. Helm does not upgrade CRDs from a chart's crds directory, so the chart ships this Job. The listing gives the rest.
   - Before deleting, remove a chart-managed GPUCluster and wait for it to go. The Job deletes the chart-managed GPUCluster object and waits until it is gone, so the operator can drain workloads under its finalizer while it is still running. The listing gives the rest.
   - After deleting, prune node-feature-discovery labels from the nodes. The Job runs nfd-master -prune, which removes node-feature-discovery labels, annotations and taints from every node, with its own ServiceAccount, ClusterRole and ClusterRoleBinding. The listing gives the rest.
EOF_NOT_RUN