From 488c4166e373338c4d39051a8eff1f4483dadde7 Mon Sep 17 00:00:00 2001
From: Berto D'Attoma <88311595+bdattoma@users.noreply.github.com>
Date: Mon, 13 Jan 2025 14:42:57 +0100
Subject: [PATCH] Backport "Update AMD Operator and NFD install scripts
 (#2139)" in releases/2.13.0  (#2169)

Update AMD Operator and NFD install scripts (#2139)

* workaround for amd certified operator in ocp < 4.16

* add NFD installation script and use it in AMD script

* use NFD install script in NVIDIA script

* minor change

* update warn msg

* rm unused function
---
 .../Provisioning/GPU/AMD/amd_operator.sh      | 31 ++++++++++++++++---
 .../Provisioning/GPU/NFD/install_nfd.sh       | 29 +++++++++++++++++
 .../GPU/{ => NFD}/nfd_deploy.yaml             |  4 +--
 .../GPU/{ => NFD}/nfd_operator.yaml           |  0
 .../Provisioning/GPU/NVIDIA/gpu_deploy.sh     |  6 ++--
 5 files changed, 60 insertions(+), 10 deletions(-)
 create mode 100644 ods_ci/tasks/Resources/Provisioning/GPU/NFD/install_nfd.sh
 rename ods_ci/tasks/Resources/Provisioning/GPU/{ => NFD}/nfd_deploy.yaml (85%)
 rename ods_ci/tasks/Resources/Provisioning/GPU/{ => NFD}/nfd_operator.yaml (100%)

diff --git a/ods_ci/tasks/Resources/Provisioning/GPU/AMD/amd_operator.sh b/ods_ci/tasks/Resources/Provisioning/GPU/AMD/amd_operator.sh
index 4ff4b4d88..aa27ef97b 100755
--- a/ods_ci/tasks/Resources/Provisioning/GPU/AMD/amd_operator.sh
+++ b/ods_ci/tasks/Resources/Provisioning/GPU/AMD/amd_operator.sh
@@ -157,6 +157,32 @@ EOF
   fi 
 }
 
+function applyWorkaroundForOlderOCPVersions () {
+  # workaround for OCP versions less than 4.16
+  # AMD certified operator is published starting from OCP v4.16
+  ocpVersion=$(oc version --output json | jq '.openshiftVersion' | tr -d '"')
+  IFS='.' read -ra ocpVersionSplit <<< "$ocpVersion"
+  if [ "${ocpVersionSplit[1]}" -lt 16 ]; then
+    echo "OCP Version: $ocpVersion"
+    echo "AMD Operator is not available for versions < 4.16, hence creating custom catalog source as workaround"
+    oc apply -f - <<EOF
+apiVersion: operators.coreos.com/v1alpha1
+kind: CatalogSource
+metadata:
+  name: certified-operators-416-amd
+  namespace: openshift-marketplace
+spec:
+  displayName: Certfied operator
+  image: 'registry.redhat.io/redhat/certified-operator-index:v4.16'
+  publisher: RHOAI QE
+  sourceType: grpc
+EOF
+    oc wait --timeout="120s" --for=condition=ready=true pod -n openshift-marketplace -l olm.catalogSource=certified-operators-416-amd
+    sed -i'' -e "s/certified-operators/certified-operators-416-amd/g" "$GPU_INSTALL_DIR/amd_gpu_install.yaml"
+  fi
+}
+
+applyWorkaroundForOlderOCPVersions
 check_registry
 status=$?
 
@@ -170,10 +196,7 @@ fi
 sleep 120
 wait_while 1800 ! machineconfig_updates
 
-echo "Installing NFD operator"
-oc apply -f "$GPU_INSTALL_DIR/../nfd_operator.yaml"
-wait_while 360 ! has_csv_succeeded openshift-nfd nfd
-oc apply -f "$GPU_INSTALL_DIR/../nfd_deploy.yaml"
+/bin/bash tasks/Resources/Provisioning/GPU/NFD/install_nfd.sh
 echo "Installing KMM operator"
 oc apply -f "$GPU_INSTALL_DIR/kmm_operator_install.yaml"
 wait_while 360 ! has_csv_succeeded openshift-kmm kernel-module-management
diff --git a/ods_ci/tasks/Resources/Provisioning/GPU/NFD/install_nfd.sh b/ods_ci/tasks/Resources/Provisioning/GPU/NFD/install_nfd.sh
new file mode 100644
index 000000000..7a7108caa
--- /dev/null
+++ b/ods_ci/tasks/Resources/Provisioning/GPU/NFD/install_nfd.sh
@@ -0,0 +1,29 @@
+#!/bin/bash
+set -e
+
+NFD_INSTALL_DIR="$(dirname "$0")"
+NFD_INSTANCE=$NFD_INSTALL_DIR/nfd_deploy.yaml
+echo "Installing NFD operator"
+oc apply -f "$NFD_INSTALL_DIR/nfd_operator.yaml"
+oc wait --timeout=3m --for jsonpath='{.status.state}'=AtLatestKnown -n openshift-nfd sub nfd
+
+ocpVersion=$(oc version --output json | jq '.openshiftVersion' | tr -d '"')
+IFS='.' read -ra ocpVersionSplit <<< "$ocpVersion"
+xyVersion="${ocpVersionSplit[0]}.${ocpVersionSplit[1]}"
+declare -A images=(
+    ["4.14"]="registry.redhat.io\/openshift4\/ose-node-feature-discovery@sha256:2977e67a413882efbfb90b52facf65d38a5cb2cd7a232ca3a69476e5dec33319"
+    ["4.15"]="registry.redhat.io\/openshift4\/ose-node-feature-discovery-rhel9@sha256:661b6697dee34626a3a98b50cdba787402ab214d2807b8460df92e3c79cdfcc5"
+    ["4.16"]="registry.redhat.io\/openshift4\/ose-node-feature-discovery-rhel9@sha256:bb95bc317ab78e8af4ef34dd66f9f62c2f8c261dfb5eab40918142812802f8b7"
+    ["4.17"]="registry.redhat.io\/openshift4\/ose-node-feature-discovery-rhel9@sha256:154cf3f1ddaf895d7ecd04947bd455a930132f72acc6e8bde8c26bc123184ace"
+    # 4.18 is a pre-release image. We need to update it later
+    ["4.18"]="registry.redhat.io\/openshift4\/ose-node-feature-discovery-rhel9@sha256:510cb4351253492455664b6c323f54dc2f6f2f8791c5e92ba6b7e60b8adb357c"
+)
+if [ "${images[$xyVersion]}" ]; then
+    imageUrl="${images[$xyVersion]}"
+    echo "Using image SHA for $xyVersion: $imageUrl"
+else
+    imageUrl="${images["4.17"]}"
+    echo "WARNING: I don't know the sha for $xyVersion. Re-using default 4.17 $imageUrl. It might not work!"
+fi
+sed -i'' -e "s/<imageUrl>/$imageUrl/g" $NFD_INSTANCE
+oc apply -f "$NFD_INSTANCE"
diff --git a/ods_ci/tasks/Resources/Provisioning/GPU/nfd_deploy.yaml b/ods_ci/tasks/Resources/Provisioning/GPU/NFD/nfd_deploy.yaml
similarity index 85%
rename from ods_ci/tasks/Resources/Provisioning/GPU/nfd_deploy.yaml
rename to ods_ci/tasks/Resources/Provisioning/GPU/NFD/nfd_deploy.yaml
index 113980150..ffa3dd1f9 100644
--- a/ods_ci/tasks/Resources/Provisioning/GPU/nfd_deploy.yaml
+++ b/ods_ci/tasks/Resources/Provisioning/GPU/NFD/nfd_deploy.yaml
@@ -7,8 +7,8 @@ spec:
   instance: "" # instance is empty by default
   topologyupdater: false # False by default
   operand:
-    # Image digest for registry.redhat.io/openshift4/ose-node-feature-discovery:v4.11
-    image: registry.redhat.io/openshift4/ose-node-feature-discovery@sha256:d6242132d2ddec00c46d22b63015a33af821eace0150ba47d185cd992fee317d
+    # Image URL example: registry.redhat.io/openshift4/ose-node-feature-discovery:v4.11
+    image: <imageUrl>
     imagePullPolicy: Always
   workerConfig:
     configData: |
diff --git a/ods_ci/tasks/Resources/Provisioning/GPU/nfd_operator.yaml b/ods_ci/tasks/Resources/Provisioning/GPU/NFD/nfd_operator.yaml
similarity index 100%
rename from ods_ci/tasks/Resources/Provisioning/GPU/nfd_operator.yaml
rename to ods_ci/tasks/Resources/Provisioning/GPU/NFD/nfd_operator.yaml
diff --git a/ods_ci/tasks/Resources/Provisioning/GPU/NVIDIA/gpu_deploy.sh b/ods_ci/tasks/Resources/Provisioning/GPU/NVIDIA/gpu_deploy.sh
index 5edbff0b3..8ccf4f290 100755
--- a/ods_ci/tasks/Resources/Provisioning/GPU/NVIDIA/gpu_deploy.sh
+++ b/ods_ci/tasks/Resources/Provisioning/GPU/NVIDIA/gpu_deploy.sh
@@ -12,10 +12,9 @@ CSVNAME="$(oc get packagemanifests/gpu-operator-certified -n openshift-marketpla
 sed -i'' -e "0,/v1.11/s//$CHANNEL/g" "$GPU_INSTALL_DIR/gpu_install.yaml"
 
 oc apply -f "$GPU_INSTALL_DIR/gpu_install.yaml"
-oc apply -f "$GPU_INSTALL_DIR/../nfd_operator.yaml"
-echo "Wait for Nvidia GPU Operator Subscription, InstallPlan and Deployment to complete"
+/bin/bash tasks/Resources/Provisioning/GPU/NFD/install_nfd.sh
 
-oc wait --timeout=3m --for jsonpath='{.status.state}'=AtLatestKnown -n openshift-nfd sub nfd
+echo "Wait for Nvidia GPU Operator Subscription, InstallPlan and Deployment to complete"
 
 oc wait --timeout=3m --for jsonpath='{.status.state}'=AtLatestKnown -n nvidia-gpu-operator sub gpu-operator-certified
 
@@ -80,7 +79,6 @@ function rerun_accelerator_migration() {
 }
 
 wait_until_pod_ready_status  "gpu-operator"
-oc apply -f "$GPU_INSTALL_DIR/../nfd_deploy.yaml"
 oc get csv -n nvidia-gpu-operator "$CSVNAME" -o jsonpath='{.metadata.annotations.alm-examples}' | jq .[0] > clusterpolicy.json
 oc apply -f clusterpolicy.json
 wait_until_pod_ready_status "nvidia-device-plugin-daemonset"