Merge pull request #117 from NVIDIA/e2etestandpushimage

enhance e2e test and publish image
NVIDIA · Sep 25, 2024 · d06fa8c · d06fa8c
2 parents c238934 + e5b4884
commit d06fa8c
Show file tree

Hide file tree

Showing 5 changed files with 155 additions and 47 deletions.
diff --git a/.github/workflows/precompiled.yaml b/.github/workflows/precompiled.yaml
@@ -25,13 +25,14 @@ jobs:
     outputs:
       driver_branch: ${{ steps.extract_driver_branch.outputs.driver_branch }}
       kernel_flavors: ${{ steps.extract_driver_branch.outputs.kernel_flavors }}
+      dist: ${{ steps.extract_driver_branch.outputs.dist }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
       - name: Read driver versions
         id: extract_driver_branch
         run: |
-          # get driver-branch
+          # get driver_branch
           DRIVER_BRANCH=("535" "550")
           driver_branch_json=$(printf '%s\n' "${DRIVER_BRANCH[@]}" | jq -R . | jq -cs .)
           echo "driver_branch=$driver_branch_json" >> $GITHUB_OUTPUT
@@ -41,13 +42,19 @@ jobs:
           kernel_flavors_json=$(printf '%s\n' "${KERNEL_FLAVORS[@]}" | jq -R . | jq -cs .)
           echo "kernel_flavors=$kernel_flavors_json" >> $GITHUB_OUTPUT
 
-  precompiled-image:
+          # get ubuntu distributions
+          DIST=("ubuntu22.04")
+          dist_json=$(printf '%s\n' "${DIST[@]}" | jq -R . | jq -cs .)
+          echo "dist=$dist_json" >> $GITHUB_OUTPUT
+
+  precompiled-build-image:
     needs: set-driver-version-matrix
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        driver-branch: ${{ fromJson(needs.set-driver-version-matrix.outputs.driver_branch) }}
+        driver_branch: ${{ fromJson(needs.set-driver-version-matrix.outputs.driver_branch) }}
         flavor: ${{ fromJson(needs.set-driver-version-matrix.outputs.kernel_flavors) }}
+        dist: ${{ fromJson(needs.set-driver-version-matrix.outputs.dist) }}
     steps:
       - uses: actions/checkout@v4
         name: Check out code
@@ -59,7 +66,7 @@ jobs:
           REPO_FULL_NAME="${{ github.repository }}"
           echo "LABEL_IMAGE_SOURCE=https://github.com/${REPO_FULL_NAME}" >> $GITHUB_ENV
 
-          GENERATE_ARTIFACTS="true"
+          GENERATE_ARTIFACTS="false"
           echo "PUSH_ON_BUILD=${GENERATE_ARTIFACTS}" >> $GITHUB_ENV
           echo "BUILD_MULTI_ARCH_IMAGES=${GENERATE_ARTIFACTS}" >> $GITHUB_ENV
 
@@ -79,10 +86,10 @@ jobs:
           VERSION: ${COMMIT_SHORT_SHA}
           BASE_TARGET: jammy
         run: |
-          make DRIVER_BRANCH=${{ matrix.driver-branch }} KERNEL_FLAVOR=${{ matrix.flavor }} build-base-${BASE_TARGET}
+          make DRIVER_BRANCH=${{ matrix.driver_branch }} KERNEL_FLAVOR=${{ matrix.flavor }} build-base-${BASE_TARGET}
 
           trap "docker rm -f base-${BASE_TARGET}-${{ matrix.flavor }}" EXIT
-          docker run -d --name base-${BASE_TARGET}-${{ matrix.flavor }} ghcr.io/nvidia/driver:base-${BASE_TARGET}-${{ matrix.flavor }}-${{ matrix.driver-branch }}
+          docker run -d --name base-${BASE_TARGET}-${{ matrix.flavor }} ghcr.io/nvidia/driver:base-${BASE_TARGET}-${{ matrix.flavor }}-${{ matrix.driver_branch }}
           # try 3 times every 10 seconds to get the file, if success exit the loop
           for i in {1..3}; do
               docker cp base-${BASE_TARGET}-${{ matrix.flavor }}:/var/kernel_version.txt kernel_version.txt && break
@@ -93,19 +100,42 @@ jobs:
           IMAGE_NAME: ghcr.io/nvidia/driver
           VERSION: ${COMMIT_SHORT_SHA}
           PRECOMPILED: "true"
-          DIST: signed_ubuntu22.04
+          DIST: signed_${{ matrix.dist }}
         run: |
           source kernel_version.txt && \
-          make DRIVER_VERSIONS=${DRIVER_VERSIONS} DRIVER_BRANCH=${{ matrix.driver-branch }} build-${DIST}-${DRIVER_VERSION}
+          make DRIVER_VERSIONS=${DRIVER_VERSIONS} DRIVER_BRANCH=${{ matrix.driver_branch }} build-${DIST}-${DRIVER_VERSION}
+
+      - name: Save build image as a tar
+        env:
+          DIST: ${{ matrix.dist }}
+          PRIVATE_REGISTRY: "ghcr.io"
+        run: |
+          source kernel_version.txt
+          docker save "${PRIVATE_REGISTRY}/nvidia/driver:${{ matrix.driver_branch }}-${KERNEL_VERSION}-${DIST}" \
+            -o  ./driver-images-${{ matrix.driver_branch }}-${KERNEL_VERSION}-${DIST}.tar
+          # set env for artifacts upload
+          echo "KERNEL_VERSION=$KERNEL_VERSION" >> $GITHUB_ENV
+          echo "DIST=$DIST" >> $GITHUB_ENV
+
+      - name: Upload build image as an artifact
+        uses: actions/upload-artifact@v4
+        with:
+           name: driver-images-${{ matrix.driver_branch }}-${{ env.KERNEL_VERSION }}-${{ env.DIST }}
+           path: ./driver-images-${{ matrix.driver_branch }}-${{ env.KERNEL_VERSION }}-${{ env.DIST }}.tar
+           retention-days: 1
 
   determine-e2e-test-matrix:
     runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        dist: ${{ fromJson(needs.set-driver-version-matrix.outputs.dist) }}
     needs:
-      - precompiled-image
+      - precompiled-build-image
       - set-driver-version-matrix
     outputs:
       matrix_values_not_empty: ${{ steps.set_kernel_version.outputs.matrix_values_not_empty }}
       matrix_values: ${{ steps.set_kernel_version.outputs.matrix_values }}
+      dist: ${{ steps.set-driver-version-matrix.outputs.dist }}
     steps:
       - name: Check out code
         uses: actions/checkout@v4
@@ -120,42 +150,27 @@ jobs:
         id: set_kernel_version
         env:
           BASE_TARGET: "jammy"
-          DIST: "ubuntu22.04"
+          DIST: ${{ matrix.dist }}
         run: |
           echo "matrix_values_not_empty=0" >> $GITHUB_OUTPUT
 
           kernel_flavors_json='${{ needs.set-driver-version-matrix.outputs.kernel_flavors }}'
-          kernel_flavors=$(echo "$kernel_flavors_json" | jq -r '.[]')
+          KERNEL_FLAVORS=($(echo "$kernel_flavors_json" | jq -r '.[]'))
           driver_branch_json='${{ needs.set-driver-version-matrix.outputs.driver_branch }}'
-          driver_branch=$(echo "$driver_branch_json" | jq -r '.[]')
-
-          kernel_versions=()
-          for kernel_flavor in $kernel_flavors; do
-            # FIXME -- remove if condition, once azure kernel upgrade starts working
-            if [[ "$kernel_flavor" == "azure" ]]; then
-              echo "skipping azure kernel testing"
-              continue
-            fi
-            for DRIVER_BRANCH in $driver_branch; do
-              source ./tests/scripts/findkernelversion.sh "$BASE_TARGET" "${kernel_flavor}" "$DRIVER_BRANCH" "$DIST"
-              if [[ "$should_continue" == true ]]; then
-                echo "matrix_values_not_empty=1" >> $GITHUB_OUTPUT
-                break
-              fi
-            done
-            if [[ "$should_continue" == false ]]; then
-                echo "Skipping e2e tests for the following driver tag: ${KERNEL_VERSION}-${kernel_flavor}-${DIST}"
-            else
-              KERNEL_VERSION=$(echo "$KERNEL_VERSION" | tr -d ' \n')
-              kernel_versions+=("$KERNEL_VERSION")
-               echo "Adding the following tag to the e2e test matrix: ${KERNEL_VERSION}-${kernel_flavor}-${DIST}"
-            fi
-          done
+          DRIVER_BRANCHES=($(echo "$driver_branch_json" | jq -r '.[]'))
 
+          source ./tests/scripts/ci-precompiled-helpers.sh
+          KERNEL_VERSIONS=($(get_kernel_versions_to_test $BASE_TARGET KERNEL_FLAVORS[@] DRIVER_BRANCHES[@] $DIST))
+          if [ -z "$KERNEL_VERSIONS" ]; then
+            # no new kernel release
+            echo "Skipping e2e tests"
+            exit 0
+          fi
           # Convert array to JSON format and assign
           echo "[]" > $GITHUB_WORKSPACE/matrix_values.json
-          printf '%s\n' "${kernel_versions[@]}" | jq -R . | jq -s . > $GITHUB_WORKSPACE/matrix_values.json
+          printf '%s\n' "${KERNEL_VERSIONS[@]}" | jq -R . | jq -s . > $GITHUB_WORKSPACE/matrix_values.json
           echo "matrix_values=$(cat $GITHUB_WORKSPACE/matrix_values.json | jq -c .)" >> $GITHUB_OUTPUT
+          echo "matrix_values_not_empty=1" >> $GITHUB_OUTPUT
 
   e2e-tests-nvidiadriver:
     runs-on: ubuntu-latest
@@ -166,9 +181,16 @@ jobs:
     strategy:
       matrix:
         kernel_version: ${{ fromJson(needs.determine-e2e-test-matrix.outputs.matrix_values) }}
+        dist: ${{ fromJson(needs.set-driver-version-matrix.outputs.dist) }}
     steps:
       - name: Check out code
         uses: actions/checkout@v4
+      - name: Login to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
       - name: Set up Holodeck
         uses: NVIDIA/[email protected]
         env:
@@ -195,6 +217,15 @@ jobs:
           echo "PRIVATE_REGISTRY=ghcr.io" >> $GITHUB_ENV
           KERNEL_VERSION="${{ matrix.kernel_version }}"
           echo "KERNEL_VERSION=$KERNEL_VERSION" >> $GITHUB_ENV
+          echo "DIST=${{ matrix.dist }}" >> $GITHUB_ENV
+          driver_branch_json="${{ needs.set-driver-version-matrix.outputs.driver_branch }}"
+          DRIVER_BRANCHES=($(echo "$driver_branch_json" | jq -r '.[]'))
+          echo "DRIVER_BRANCHES=${DRIVER_BRANCHES[*]}" >> $GITHUB_ENV
+
+      - name: Install GitHub CLI
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y gh
 
       - name: Upgrade the kernel for Precompiled e2e test
         env:
@@ -220,23 +251,29 @@ jobs:
       - name: Precompiled e2e test gpu driver validation
         env:
           TEST_CASE: "./tests/cases/nvidia-driver.sh"
-          GPU_OPERATOR_OPTIONS: "--set driver.repository=${{ env.PRIVATE_REGISTRY }}/nvidia --set driver.usePrecompiled=true"
+          GPU_OPERATOR_OPTIONS: "--set driver.repository=${{ env.PRIVATE_REGISTRY }}/nvidia --set driver.usePrecompiled=true \
+            --set driver.imagePullPolicy=Never"
+          GH_TOKEN:  ${{ secrets.GITHUB_TOKEN }}
         run: |
           rc=0
           # for precompiled driver we are setting driver branch as driver version
-          driver_versions_json='${{ needs.set-driver-version-matrix.outputs.driver_branch }}'
-          driver_versions=$(echo "$driver_versions_json" | jq -r '.[]')
-          for DRIVER_VERSION in $driver_versions; do
+          DRIVER_BRANCHES=(${{ env.DRIVER_BRANCHES }})
+          for DRIVER_VERSION in "${DRIVER_BRANCHES[@]}"; do
             echo "Running e2e for DRIVER_VERSION=$DRIVER_VERSION"
+            image="driver-images-${DRIVER_VERSION}-${KERNEL_VERSION}-${DIST}"
+            echo "Downloading  $image in tests directory"
+            gh run download --name $image --dir ./tests/
             status=0
-            OPERATOR_OPTIONS="${GPU_OPERATOR_OPTIONS} --set driver.version=${DRIVER_VERSION}"
+            TEST_CASE_ARGS="${GPU_OPERATOR_OPTIONS} --set driver.version=${DRIVER_VERSION}"
             # add escape character for space
-            OPERATOR_OPTIONS=$(printf '%q ' "$OPERATOR_OPTIONS")
-            ./tests/ci-run-e2e.sh "${TEST_CASE}" "${OPERATOR_OPTIONS}" || status=$?
+            TEST_CASE_ARGS=$(printf '%q ' "$TEST_CASE_ARGS")
+            IMAGE_PATH="./tests/driver-images-${DRIVER_VERSION}-${KERNEL_VERSION}-${DIST}.tar"
+            ./tests/ci-run-e2e.sh "${TEST_CASE}" "${TEST_CASE_ARGS}" ${IMAGE_PATH} || status=$?
             if [ $status -eq 1 ]; then
               echo "e2e validation failed for driver version $DRIVER_VERSION with status $status"
               rc=$status
             fi
+            rm -f $IMAGE_PATH
           done
           ./tests/scripts/pull.sh /tmp/logs logs
           exit $rc
@@ -248,3 +285,42 @@ jobs:
           name: nvidiadriver-Precompiled-e2e-test-logs
           path: ./logs/
           retention-days: 15
+
+  publish-precompiled-image:
+    runs-on: ubuntu-latest
+    needs:
+      - set-driver-version-matrix
+      - determine-e2e-test-matrix
+      - e2e-tests-nvidiadriver
+    strategy:
+      matrix:
+        driver_branch: ${{ fromJson(needs.set-driver-version-matrix.outputs.driver_branch) }}
+        kernel_version: ${{ fromJson(needs.determine-e2e-test-matrix.outputs.matrix_values) }}
+        dist: ${{ fromJson(needs.set-driver-version-matrix.outputs.dist) }}     
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Login to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name:  Set image vars
+        run: |
+          echo "PRIVATE_REGISTRY=ghcr.io" >> $GITHUB_ENV
+          echo "DIST=${{ matrix.dist }}" >> $GITHUB_ENV
+
+      - name: Download built image artifact
+        uses: actions/download-artifact@v4
+        with:
+           name: driver-images-${{ matrix.driver_branch }}-${{ matrix.kernel_version }}-${{ env.DIST }}
+           path: ./
+
+      - name: Publish image
+        run: |
+          image_path="./driver-images-${{ matrix.driver_branch }}-${{ matrix.kernel_version }}-${{ env.DIST }}.tar"
+          echo "uploading  $image_path"
+          docker load -i $image_path
+          docker push ${PRIVATE_REGISTRY}/nvidia/driver:${{ matrix.driver_branch }}-${{ matrix.kernel_version }}-${{ env.DIST }}
diff --git a/tests/cases/nvidia-driver.sh b/tests/cases/nvidia-driver.sh
@@ -8,6 +8,10 @@ fi
 
 # export gpu-operator options
 export TEST_CASE_ARGS="$1"
+if [[ $# -eq 2 ]]; then
+	export IMAGE_PATH="$2"
+	sudo ctr -n k8s.io images import "$IMAGE_PATH"
+fi
 
 SCRIPTS_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )"/../scripts && pwd )"
 source "${SCRIPTS_DIR}"/.definitions.sh

diff --git a/tests/ci-run-e2e.sh b/tests/ci-run-e2e.sh
@@ -2,7 +2,7 @@
 
 set -xe
 
-if [[ $# -ne 2 ]]; then
+if [[ $# -lt 2 ]]; then
 	echo "TEST_CASE TEST_CASE_ARGS are required"
 	exit 1
 fi

diff --git a/tests/scripts/ci-precompiled-helpers.sh b/tests/scripts/ci-precompiled-helpers.sh
@@ -0,0 +1,30 @@
+get_kernel_versions_to_test() {
+    if [[ "$#" -ne 4 ]]; then
+	    echo " Error:$0 must be called with BASE_TARGET DRIVER_BRANCHES DRIVER_BRANCHES DIST" >&2
+	    exit 1
+    fi
+
+    local BASE_TARGET="$1"
+    local -a KERNEL_FLAVORS=("${!2}")
+    local -a DRIVER_BRANCHES=("${!3}")
+    local DIST="$4"
+
+    kernel_versions=()
+    for kernel_flavor in "${KERNEL_FLAVORS[@]}"; do
+        # FIXME -- remove if condition, once azure kernel upgrade starts working
+        if [[ "$kernel_flavor" == "azure" ]]; then
+            continue
+        fi
+        for DRIVER_BRANCH in "${DRIVER_BRANCHES[@]}"; do
+            source ./tests/scripts/findkernelversion.sh "$BASE_TARGET" "${kernel_flavor}" "$DRIVER_BRANCH" "$DIST" >&2
+            if [[ "$should_continue" == true ]]; then
+                break
+            fi
+        done
+        if [[ "$should_continue" == true ]]; then
+            KERNEL_VERSION=$(echo "$KERNEL_VERSION" | tr -d ' \n')
+            kernel_versions+=("$KERNEL_VERSION")
+        fi
+    done
+    echo "${kernel_versions[@]}"
+}
diff --git a/tests/scripts/findkernelversion.sh b/tests/scripts/findkernelversion.sh
@@ -10,7 +10,7 @@ export KERNEL_FLAVOR="${2}"
 export DRIVER_BRANCH="${3}"
 export DIST="${4}"
 
-export REGCTL_VERSION=v0.4.7
+export REGCTL_VERSION=v0.7.1
 mkdir -p bin
 curl -sSLo bin/regctl https://github.com/regclient/regclient/releases/download/${REGCTL_VERSION}/regctl-linux-amd64
 chmod a+x bin/regctl
@@ -22,8 +22,6 @@ export $(grep -oP 'KERNEL_VERSION=[^ ]+' ./kernel_version.txt)
 
 # calculate driver tag
 status=0
-echo "regctl tag ls  nvcr.io/nvidia/driver | grep "^${DRIVER_BRANCH}-${KERNEL_VERSION}-${DIST}$""
-
 regctl tag ls  nvcr.io/nvidia/driver | grep "^${DRIVER_BRANCH}-${KERNEL_VERSION}-${DIST}$" || status=$?
 if [[ $status -eq 0 ]]; then
     export should_continue=false
-Original file line number
+Diff line change
@@ Expand Up / @@ -2,7 +2,7 @@ @@
     set -xe
-    if [[ $# -ne 2 ]]; then
+    if [[ $# -lt 2 ]]; then
     	echo "TEST_CASE TEST_CASE_ARGS are required"
     	exit 1
     fi
@@ Expand Down @@