Test Not for review holodeck kernel version change

Signed-off-by: shiva kumar <shivaku@nvidia.com>

Author: shivakunv

.github/workflows/image.yaml

@@ -21,12 +21,12 @@ on:
       - opened
       - synchronize
     branches:
-      - main
-      - release-*
+      - main-no
+      - release-no
   push:
     branches:
-      - main
-      - release-*
+      - main-no
+      - release-no
 
 jobs:
   image:

.github/workflows/precompiled.yaml

@@ -16,8 +16,15 @@
 name: Precompiled images
 
 on:
-  schedule:
-    - cron: '00 09 * * *'
+  pull_request:
+    types:
+      - opened
+      - synchronize
+    branches:
+      - test-holodeck
+  push:
+    branches:
+      - test-holodeck
 
 jobs:
   set-driver-version-matrix:
@@ -39,17 +46,20 @@ jobs:
           echo "driver_branch=$driver_branch_json" >> $GITHUB_OUTPUT
 
           # get kernel flavors
-          KERNEL_FLAVORS=("aws" "azure" "generic" "nvidia" "oracle")
+          # KERNEL_FLAVORS=("aws" "azure" "generic" "nvidia" "oracle")
+          KERNEL_FLAVORS=("nvidia")
           kernel_flavors_json=$(printf '%s\n' "${KERNEL_FLAVORS[@]}" | jq -R . | jq -cs .)
           echo "kernel_flavors=$kernel_flavors_json" >> $GITHUB_OUTPUT
 
           # get ubuntu distributions
-          DIST=("ubuntu22.04" "ubuntu24.04")
+          # DIST=("ubuntu22.04" "ubuntu24.04")
+          DIST=("ubuntu22.04")
           dist_json=$(printf '%s\n' "${DIST[@]}" | jq -R . | jq -cs .)
           echo "dist=$dist_json" >> $GITHUB_OUTPUT
 
           # LTS_KERNEL setup
-          LTS_KERNEL=("5.15" "6.8")
+          # LTS_KERNEL=("5.15" "6.8")
+          LTS_KERNEL=("6.8")
           lts_kernel_json=$(printf '%s\n' "${LTS_KERNEL[@]}" | jq -R . | jq -cs .)
           echo "lts_kernel=$lts_kernel_json" >> $GITHUB_OUTPUT
 
@@ -303,8 +313,12 @@ jobs:
           DRIVER_BRANCHES=($(echo "$driver_branch_json" | jq -r '.[]'))
           echo "DRIVER_BRANCHES=${DRIVER_BRANCHES[*]}" >> $GITHUB_ENV
 
+      - name: Set kernel version in holodeck_${{ env.DIST }}.yaml
+        run: |
+          yq e '.spec.kernel.version = "${{ env.KERNEL_VERSION }}"' -i tests/holodeck_${{ env.DIST }}.yaml
+
       - name: Set up Holodeck
-        uses: NVIDIA/holodeck@v0.2.6
+        uses: NVIDIA/holodeck@main
         env:
           AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
           AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
@@ -329,27 +343,6 @@ jobs:
           sudo apt-get update
           sudo apt-get install -y gh
 
-      - name: Upgrade the kernel for Precompiled e2e test
-        env:
-          UPGRADE_KERNEL_SCRIPT: "./tests/scripts/upgrade-kernel.sh"
-        run: |
-          status=0
-          ./tests/ci-remote-exec.sh "${UPGRADE_KERNEL_SCRIPT}" "${KERNEL_VERSION}" || status=$?
-          # On the target system, all scripts/test-case exit with code 1 for error handling.
-          # However, since reboot-related disconnections break the SSH connection
-          # and can cause the entire job to exit, we should ignore all errors except
-          # exit code 1. During a reboot, exit code 1 will not be thrown, so handling
-          # other errors as code 1 will ensure proper management of reboot scenarios
-          if [ $status -eq 1 ]; then
-            echo "Kernel version $KERNEL_VERSION upgrade failed"
-            exit 1
-          fi
-          ./tests/scripts/remote_retry.sh || status=$?
-          if [ $status -ne 0 ]; then
-              echo "Failed to connect to remote instance"
-              exit $status
-          fi
-
       - name: Precompiled e2e test gpu driver validation
         env:
           TEST_CASE: "./tests/cases/nvidia-driver.sh"
@@ -376,6 +369,11 @@ jobs:
             # add escape character for space
             TEST_CASE_ARGS=$(printf '%q ' "$TEST_CASE_ARGS")
             IMAGE_PATH="./tests/driver-images-${DRIVER_VERSION}-${KERNEL_VERSION}-${DIST}.tar"
+            # ./tests/scripts/remote_retry.sh || status=$?
+            # if [ $status -ne 0 ]; then
+            #   echo "Failed to connect to remote instance"
+            #   exit $status
+            # fi
             ./tests/ci-run-e2e.sh "${TEST_CASE}" "${TEST_CASE_ARGS}" ${IMAGE_PATH} || status=$?
             if [ $status -eq 1 ]; then
               echo "e2e validation failed for driver version $DRIVER_VERSION with status $status"

tests/holodeck_ubuntu22.04.yaml

@@ -27,8 +27,11 @@ spec:
   containerRuntime:
     install: true
     name: containerd
+    version: 1.6.27
   kubernetes:
     install: true
     installer: kubeadm
-    version: v1.28.5
-    crictlVersion: v1.28.0
+    version: v1.29.3
+    crictlVersion: v1.29.0
+  kernel:
+    version: 5.15.0-139-generic

tests/holodeck_ubuntu24.04.yaml

@@ -31,5 +31,7 @@ spec:
   kubernetes:
     install: true
     installer: kubeadm
-    version: v1.30.0
-    crictlVersion: v1.30.0
+    version: v1.28.5
+    crictlVersion: v1.28.0
+  kernel:
+    version: 6.8.0-59-generic

tests/scripts/.definitions.sh

@@ -16,7 +16,7 @@ CASES_DIR="$( cd "${TEST_DIR}/cases" && pwd )"
 
 : ${HELM_NVIDIA_REPO:="https://helm.ngc.nvidia.com/nvidia"}
 
-: ${DAEMON_POD_STATUS_TIME_OUT:="15m"}
+: ${DAEMON_POD_STATUS_TIME_OUT:="30m"}
 : ${POD_STATUS_TIME_OUT:="2m"}
 
 : ${LOG_DIR:="/tmp/logs"}

tests/scripts/findkernelversion.sh

@@ -40,3 +40,5 @@ if [[ $status -eq 0 ]]; then
 else
     export should_continue=true
 fi
+# SHIVA remove this
+export should_continue=true