[no-relnote] Update Github Actions E2E

Signed-off-by: Carlos Eduardo Arango Gutierrez <[email protected]>

Author: ArangoGutierrez

.github/workflows/e2e.yaml

@@ -70,8 +70,8 @@ jobs:
 
       - name: Run e2e tests
         env:
-          IMAGE_NAME: ghcr.io/nvidia/container-toolkit
-          VERSION: ${{ inputs.version }}
+          E2E_IMAGE_REPO: ghcr.io/nvidia/container-toolkit
+          E2E_IMAGE_TAG: ${{ inputs.version }}-ubuntu20.04
           SSH_KEY: ${{ secrets.AWS_SSH_KEY }}
           E2E_SSH_USER: ${{ secrets.E2E_SSH_USER }}
           E2E_SSH_HOST: ${{ steps.holodeck_public_dns_name.outputs.result }}
@@ -84,6 +84,13 @@ jobs:
 
           make -f tests/e2e/Makefile test
 
+      - name: Archive Ginkgo logs
+        uses: actions/upload-artifact@v4
+        with:
+          name: ginkgo-logs
+          path: ginkgo.json
+          retention-days: 15
+        
       - name: Send Slack alert notification
         if: ${{ failure() }}
         uses: slackapi/[email protected]

.gitignore

@@ -11,3 +11,4 @@
 /nvidia-ctk
 /shared-*
 /release-*
+/bin

tests/e2e/Makefile

@@ -13,14 +13,16 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-.PHONY: test-e2e ginkgo
+.PHONY: test $(GINKGO_BIN)
 
 GINKGO_ARGS ?=
 LOG_ARTIFACTS_DIR ?= $(CURDIR)/e2e_logs
 
-ginkgo:
+GINKGO_BIN := $(CURDIR)/bin/ginkgo
+
+test: $(GINKGO_BIN)
+	$(GINKGO_BIN) $(GINKGO_ARGS) -v --json-report ginkgo.json ./tests/e2e/...
+
+$(GINKGO_BIN):
 	mkdir -p $(CURDIR)/bin
 	GOBIN=$(CURDIR)/bin go install github.com/onsi/ginkgo/v2/ginkgo@latest
-
-test-e2e: ginkgo
-	$(CURDIR)/bin/ginkgo $(GINKGO_ARGS) -v --json-report ginkgo.json ./tests/e2e/...

tests/e2e/README.md

@@ -20,7 +20,7 @@ limitations under the License.
 ---
 
 ## 1  Scope & Goals
-This repository contains a **Ginkgo v2 / Gomega** test harness that exercises an
+This folder contains a **Ginkgo v2 / Gomega** test harness that exercises an
 NVIDIA Container Toolkit (CTK) installation on a **remote GPU‑enabled host** via
 SSH.  The suite validates that:
 
@@ -58,12 +58,13 @@ compatibility runs, and pre‑release validation of new CTK builds.
 
 | Variable | Required | Example | Description |
 |----------|----------|---------|-------------|
-| `INSTALL_CTK` | ✖ | `true` | When `true` the test installs CTK on the remote host before running the image. When `false` it assumes CTK is already present. |
-| `TOOLKIT_IMAGE` | ✔ | `nvcr.io/nvidia/cuda:12.4.0-runtime-ubi9` | Image that will be pulled & executed. |
-| `SSH_KEY` | ✔ | `/home/ci/.ssh/id_rsa` | Private key used for authentication. |
-| `SSH_USER` | ✔ | `ubuntu` | Username on the remote host. |
-| `REMOTE_HOST` | ✔ | `gpurunner01.corp.local` | Hostname or IP address of the target node. |
-| `REMOTE_PORT` | ✔ | `22` | SSH port of the target node. |
+| `E2E_INSTALL_CTK` | ✖ | `true` | When `true` the test installs CTK on the remote host before running the image. When `false` it assumes CTK is already present. |
+| `E2E_IMAGE_REPO` | ✔ | `ghcr.io/nvidia/container-toolkit` | Container Toolkit Image  |
+| `E2E_IMAGE_TAG` | ✔ | `latest` | Image tag |
+| `E2E_SSH_KEY` | ✔ | `/home/ci/.ssh/id_rsa` | Private key used for authentication. |
+| `E2E_SSH_USER` | ✔ | `ubuntu` | Username on the remote host. |
+| `E2E_SSH_HOST` | ✔ | `10.0.0.0` | Hostname or IP address of the target node. |
+| `E2E_SSH_PORT` | ✔ | `22` | SSH port of the target node. |
 
 > All variables are validated at start‑up; the suite aborts early with a clear
 > message if any are missing or ill‑formed.
@@ -87,24 +88,7 @@ bin/ginkgo:
 
 ---
 
-## 6  Running the suite
-
-### 6.1  Basic invocation
-```bash
-INSTALL_CTK=true \
-TOOLKIT_IMAGE=nvcr.io/nvidia/cuda:12.4.0-runtime-ubi9 \
-SSH_KEY=$HOME/.ssh/id_rsa \
-SSH_USER=ubuntu \
-REMOTE_HOST=10.0.0.15 \
-REMOTE_PORT=22 \
-make test-e2e
-```
-This downloads the image on the remote host, installs CTK (if requested), and
-executes a minimal CUDA‑based workload.
-
----
-
-## 7  Internal test flow
+## 6  Internal test flow
 
 | Phase | Key function(s) | Notes |
 |-------|-----------------|-------|
@@ -116,7 +100,7 @@ executes a minimal CUDA‑based workload.
 
 ---
 
-## 8  Extending the suite
+## 7  Extending the suite
 
 1. Create a new `_test.go` file under `tests/e2e`.
 2. Use the Ginkgo DSL (`Describe`, `When`, `It` …). Each leaf node receives a
@@ -127,14 +111,14 @@ executes a minimal CUDA‑based workload.
 
 ---
 
-## 9  Common issues & fixes
+## 8  Common issues & fixes
 
 | Symptom | Likely cause | Fix |
 |---------|--------------|-----|
 | `Permission denied (publickey)` | Wrong `SSH_KEY` or `SSH_USER` | Check variables; ensure key is readable by the CI user. |
 | `docker: Error response from daemon: could not select device driver` | CTK not installed or wrong runtime class | Verify `INSTALL_CTK=true` or confirm CTK installation on the host. |
 | Test hangs at image pull | No outbound internet on remote host | Pre‑load the image or use a local registry mirror. |
 
-## 10  License
+## 9  License
 Distributed under the terms of the **Apache License 2.0** (see header).
 

tests/e2e/e2e_test.go

@@ -1,6 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: Apache-2.0
+ * Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -19,9 +18,8 @@ package e2e
 
 import (
 	"context"
+	"errors"
 	"os"
-	"path/filepath"
-	"runtime"
 	"strconv"
 	"testing"
 
@@ -38,12 +36,11 @@ var (
 	ImageRepo string
 	ImageTag  string
 
-	sshKey      string
-	sshUser     string
-	host        string
-	sshPort     string
-	cwd         string
-	packagePath string
+	sshKey  string
+	sshUser string
+	sshHost string
+	sshPort string
+	cwd     string
 
 	runner Runner
 )
@@ -64,7 +61,7 @@ func TestMain(t *testing.T) {
 // BeforeSuite runs before the test suite
 var _ = BeforeSuite(func() {
 	runner = NewRunner(
-		WithHost(host),
+		WithHost(sshHost),
 		WithPort(sshPort),
 		WithSshKey(sshKey),
 		WithSshUser(sshUser),
@@ -81,59 +78,70 @@ var _ = BeforeSuite(func() {
 		err = installer.Install()
 		Expect(err).ToNot(HaveOccurred())
 	}
-
-	_, _, err := runner.Run("docker pull ubuntu")
-	Expect(err).ToNot(HaveOccurred())
-
-	_, _, err = runner.Run("docker pull nvcr.io/nvidia/k8s/cuda-sample:vectoradd-cuda12.5.0")
-	Expect(err).ToNot(HaveOccurred())
-
-	_, _, err = runner.Run("docker pull nvcr.io/nvidia/cuda:12.8.0-base-ubi8")
-	Expect(err).ToNot(HaveOccurred())
 })
 
 // getTestEnv gets the test environment variables
 func getTestEnv() {
 	defer GinkgoRecover()
 	var err error
 
-	_, thisFile, _, _ := runtime.Caller(0)
-	packagePath = filepath.Dir(thisFile)
+	installCTK = getEnvVarOrDefault("E2E_INSTALL_CTK", true)
 
-	installCTK = getBoolEnvVar("INSTALL_CTK", false)
-
-	ImageRepo = os.Getenv("E2E_IMAGE_REPO")
-	Expect(ImageRepo).NotTo(BeEmpty(), "E2E_IMAGE_REPO environment variable must be set")
+	if installCTK {
+		ImageRepo = os.Getenv("E2E_IMAGE_REPO")
+		Expect(ImageRepo).NotTo(BeEmpty(), "E2E_IMAGE_REPO environment variable must be set")
 
-	ImageTag = os.Getenv("E2E_IMAGE_TAG")
-	Expect(ImageTag).NotTo(BeEmpty(), "E2E_IMAGE_TAG environment variable must be set")
+		ImageTag = os.Getenv("E2E_IMAGE_TAG")
+		Expect(ImageTag).NotTo(BeEmpty(), "E2E_IMAGE_TAG environment variable must be set")
+	}
 
-	sshKey = os.Getenv("SSH_KEY")
-	Expect(sshKey).NotTo(BeEmpty(), "SSH_KEY environment variable must be set")
+	sshKey = os.Getenv("E2E_SSH_KEY")
+	Expect(sshKey).NotTo(BeEmpty(), "E2E_SSH_KEY environment variable must be set")
 
-	sshUser = os.Getenv("SSH_USER")
-	Expect(sshUser).NotTo(BeEmpty(), "SSH_USER environment variable must be set")
+	sshUser = os.Getenv("E2E_SSH_USER")
+	Expect(sshUser).NotTo(BeEmpty(), "E2E_SSH_USER environment variable must be set")
 
-	host = os.Getenv("REMOTE_HOST")
-	Expect(host).NotTo(BeEmpty(), "REMOTE_HOST environment variable must be set")
+	sshHost = os.Getenv("E2E_SSH_HOST")
+	Expect(sshHost).NotTo(BeEmpty(), "E2E_SSH_HOST environment variable must be set")
 
-	sshPort = os.Getenv("REMOTE_PORT")
-	Expect(sshPort).NotTo(BeEmpty(), "REMOTE_PORT environment variable must be set")
+	sshPort = getEnvVarOrDefault("E2E_SSH_PORT", "22")
 
 	// Get current working directory
 	cwd, err = os.Getwd()
 	Expect(err).NotTo(HaveOccurred())
 }
 
-// getBoolEnvVar returns the boolean value of the environment variable or the default value if not set.
-func getBoolEnvVar(key string, defaultValue bool) bool {
+func getEnvVarAs[T any](key string) (T, error) {
+	var zero T
 	value := os.Getenv(key)
 	if value == "" {
-		return defaultValue
+		return zero, errors.New("env var not set")
+	}
+
+	switch any(zero).(type) {
+	case bool:
+		v, err := strconv.ParseBool(value)
+		if err != nil {
+			return zero, err
+		}
+		return any(v).(T), nil
+	case int:
+		v, err := strconv.Atoi(value)
+		if err != nil {
+			return zero, err
+		}
+		return any(v).(T), nil
+	case string:
+		return any(value).(T), nil
+	default:
+		return zero, errors.New("unsupported type")
 	}
-	boolValue, err := strconv.ParseBool(value)
+}
+
+func getEnvVarOrDefault[T any](key string, defaultValue T) T {
+	val, err := getEnvVarAs[T](key)
 	if err != nil {
 		return defaultValue
 	}
-	return boolValue
+	return val
 }

tests/e2e/installer.go

@@ -1,6 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
- * SPDX-License-Identifier: Apache-2.0
+ * Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -14,6 +13,7 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
+
 package e2e
 
 import (