Merge pull request #768 from elezar/add-toolkit-install-unit-tests

elezar · web-flow · commit ab0a4eaa1980 · 2024-11-06T00:41:55.000+01:00
[no-relnote] Add toolkit install unit test
diff --git a/.gitignore b/.gitignore
@@ -1,5 +1,5 @@
-dist
-artifacts
+/dist
+/artifacts
 *.swp
 *.swo
 /coverage.out*
@@ -10,4 +10,4 @@ artifacts
 /nvidia-container-toolkit
 /nvidia-ctk
 /shared-*
-/release-*
+/release-*
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
@@ -176,12 +176,6 @@ image-packaging:
       optional: true
 
 # Define publish test helpers
-.test:toolkit:
-  extends:
-    - .integration
-  variables:
-    TEST_CASES: "toolkit"
-
 .test:docker:
   extends:
     - .integration
diff --git a/deployments/container/Makefile b/deployments/container/Makefile
@@ -122,7 +122,7 @@ build-packaging: PACKAGE_DIST = all
 # Test targets
 test-%: DIST = $(*)
 
-TEST_CASES ?= toolkit docker crio containerd
+TEST_CASES ?= docker crio containerd
 $(TEST_TARGETS): test-%:
 	TEST_CASES="$(TEST_CASES)" bash -x $(CURDIR)/test/container/main.sh run \
 		$(CURDIR)/shared-$(*) \
diff --git a/internal/lookup/device.go b/internal/lookup/device.go
@@ -28,9 +28,15 @@ const (
 // NewCharDeviceLocator creates a Locator that can be used to find char devices at the specified root. A logger is
 // also specified.
 func NewCharDeviceLocator(opts ...Option) Locator {
+	filter := assertCharDevice
+	// TODO: We should have a better way to inject this logic than this envvar.
+	if os.Getenv("__NVCT_TESTING_DEVICES_ARE_FILES") == "true" {
+		filter = assertFile
+	}
+
 	opts = append(opts,
 		WithSearchPaths("", devRoot),
-		WithFilter(assertCharDevice),
+		WithFilter(filter),
 	)
 	return NewFileLocator(
 		opts...,
diff --git a/test/container/main.sh b/test/container/main.sh
@@ -19,7 +19,6 @@ shopt -s lastpipe
 readonly basedir="$(dirname "$(realpath "$0")")"
 source "${basedir}/common.sh"
 
-source "${basedir}/toolkit_test.sh"
 source "${basedir}/docker_test.sh"
 source "${basedir}/crio_test.sh"
 source "${basedir}/containerd_test.sh"
@@ -66,7 +65,7 @@ done
 
 trap '"$CLEANUP" && testing::cleanup' ERR
 
-readonly test_cases="${TEST_CASES:-toolkit docker crio containerd}"
+readonly test_cases="${TEST_CASES:-docker crio containerd}"
 
 testing::cleanup
 for tc in ${test_cases}; do
diff --git a/test/container/toolkit_test.sh b/test/container/toolkit_test.sh
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-cdi-hook b/testdata/installer/artifacts/deb/usr/bin/nvidia-cdi-hook
@@ -0,0 +1 @@
+nvidia-cdi-hook
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-container-cli b/testdata/installer/artifacts/deb/usr/bin/nvidia-container-cli
@@ -0,0 +1 @@
+nvidia-container-cli
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime b/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime
@@ -0,0 +1 @@
+nvidia-container-runtime
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime-hook b/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime-hook
@@ -0,0 +1 @@
+nvidia-container-runtime-hook
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime.cdi b/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime.cdi
@@ -0,0 +1 @@
+nvidia-container-runtime.cdi
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime.legacy b/testdata/installer/artifacts/deb/usr/bin/nvidia-container-runtime.legacy
@@ -0,0 +1 @@
+nvidia-container-runtime.legacy
diff --git a/testdata/installer/artifacts/deb/usr/bin/nvidia-ctk b/testdata/installer/artifacts/deb/usr/bin/nvidia-ctk
@@ -0,0 +1 @@
+nvidia-ctk
diff --git a/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container-go.so b/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container-go.so
@@ -0,0 +1 @@
+libnvidia-container-go.so.1
diff --git a/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container-go.so.1 b/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container-go.so.1
@@ -0,0 +1 @@
+libnvidia-container-go.so.99.88.77
diff --git a/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container-go.so.99.88.77 b/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container-go.so.99.88.77
diff --git a/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container.so b/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container.so
@@ -0,0 +1 @@
+libnvidia-container.so.1
diff --git a/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container.so.1 b/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container.so.1
@@ -0,0 +1 @@
+libnvidia-container.so.99.88.77
diff --git a/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container.so.99.88.77 b/testdata/installer/artifacts/deb/usr/lib/x86_64-linux-gnu/libnvidia-container.so.99.88.77
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-cdi-hook b/testdata/installer/artifacts/rpm/usr/bin/nvidia-cdi-hook
@@ -0,0 +1 @@
+nvidia-cdi-hook
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-cli b/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-cli
@@ -0,0 +1 @@
+nvidia-container-cli
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime b/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime
@@ -0,0 +1 @@
+nvidia-container-runtime
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime-hook b/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime-hook
@@ -0,0 +1 @@
+nvidia-container-runtime-hook
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime.cdi b/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime.cdi
@@ -0,0 +1 @@
+nvidia-container-runtime.cdi
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime.legacy b/testdata/installer/artifacts/rpm/usr/bin/nvidia-container-runtime.legacy
@@ -0,0 +1 @@
+nvidia-container-runtime.legacy
diff --git a/testdata/installer/artifacts/rpm/usr/bin/nvidia-ctk b/testdata/installer/artifacts/rpm/usr/bin/nvidia-ctk
@@ -0,0 +1 @@
+nvidia-ctk
diff --git a/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container-go.so b/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container-go.so
@@ -0,0 +1 @@
+libnvidia-container-go.so.1
diff --git a/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container-go.so.1 b/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container-go.so.1
@@ -0,0 +1 @@
+libnvidia-container-go.so.99.88.77
diff --git a/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container-go.so.99.88.77 b/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container-go.so.99.88.77
diff --git a/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container.so b/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container.so
@@ -0,0 +1 @@
+libnvidia-container.so.1
diff --git a/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container.so.1 b/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container.so.1
@@ -0,0 +1 @@
+libnvidia-container.so.99.88.77
diff --git a/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container.so.99.88.77 b/testdata/installer/artifacts/rpm/usr/lib64/libnvidia-container.so.99.88.77
diff --git a/testdata/lookup/rootfs-1/dev/nvidia0 b/testdata/lookup/rootfs-1/dev/nvidia0
diff --git a/testdata/lookup/rootfs-1/dev/nvidiactl b/testdata/lookup/rootfs-1/dev/nvidiactl
diff --git a/tools/container/nvidia-toolkit/run.go b/tools/container/nvidia-toolkit/run.go
@@ -164,7 +164,7 @@ func Run(c *cli.Context, o *options) error {
 
 		o.toolkitOptions.ContainerRuntimeRuntimes = *cli.NewStringSlice(lowlevelRuntimePaths...)
 	}
-	err = toolkit.Install(c, &o.toolkitOptions, o.toolkitRoot())
+	err = toolkit.Install(c, &o.toolkitOptions, "", o.toolkitRoot())
 	if err != nil {
 		return fmt.Errorf("unable to install toolkit: %v", err)
 	}
diff --git a/tools/container/toolkit/runtime.go b/tools/container/toolkit/runtime.go
@@ -29,10 +29,10 @@ const (
 
 // installContainerRuntimes sets up the NVIDIA container runtimes, copying the executables
 // and implementing the required wrapper
-func installContainerRuntimes(toolkitDir string, driverRoot string) error {
+func installContainerRuntimes(sourceRoot string, toolkitDir string) error {
 	runtimes := operator.GetRuntimes()
 	for _, runtime := range runtimes {
-		r := newNvidiaContainerRuntimeInstaller(runtime.Path)
+		r := newNvidiaContainerRuntimeInstaller(filepath.Join(sourceRoot, runtime.Path))
 
 		_, err := r.install(toolkitDir)
 		if err != nil {
diff --git a/tools/container/toolkit/toolkit.go b/tools/container/toolkit/toolkit.go
@@ -285,8 +285,9 @@ func TryDelete(cli *cli.Context, toolkitRoot string) error {
 }
 
 // Install installs the components of the NVIDIA container toolkit.
+// The specified sourceRoot is searched for the components to install.
 // Any existing installation is removed.
-func Install(cli *cli.Context, opts *Options, toolkitRoot string) error {
+func Install(cli *cli.Context, opts *Options, sourceRoot string, toolkitRoot string) error {
 	log.Infof("Installing NVIDIA container toolkit to '%v'", toolkitRoot)
 
 	log.Infof("Removing existing NVIDIA container toolkit installation")
@@ -307,42 +308,42 @@ func Install(cli *cli.Context, opts *Options, toolkitRoot string) error {
 		log.Errorf("Ignoring error: %v", fmt.Errorf("could not create required directories: %v", err))
 	}
 
-	err = installContainerLibraries(toolkitRoot)
+	err = installContainerLibraries(sourceRoot, toolkitRoot)
 	if err != nil && !opts.ignoreErrors {
 		return fmt.Errorf("error installing NVIDIA container library: %v", err)
 	} else if err != nil {
 		log.Errorf("Ignoring error: %v", fmt.Errorf("error installing NVIDIA container library: %v", err))
 	}
 
-	err = installContainerRuntimes(toolkitRoot, opts.DriverRoot)
+	err = installContainerRuntimes(sourceRoot, toolkitRoot)
 	if err != nil && !opts.ignoreErrors {
 		return fmt.Errorf("error installing NVIDIA container runtime: %v", err)
 	} else if err != nil {
 		log.Errorf("Ignoring error: %v", fmt.Errorf("error installing NVIDIA container runtime: %v", err))
 	}
 
-	nvidiaContainerCliExecutable, err := installContainerCLI(toolkitRoot)
+	nvidiaContainerCliExecutable, err := installContainerCLI(sourceRoot, toolkitRoot)
 	if err != nil && !opts.ignoreErrors {
 		return fmt.Errorf("error installing NVIDIA container CLI: %v", err)
 	} else if err != nil {
 		log.Errorf("Ignoring error: %v", fmt.Errorf("error installing NVIDIA container CLI: %v", err))
 	}
 
-	nvidiaContainerRuntimeHookPath, err := installRuntimeHook(toolkitRoot, toolkitConfigPath)
+	nvidiaContainerRuntimeHookPath, err := installRuntimeHook(sourceRoot, toolkitRoot, toolkitConfigPath)
 	if err != nil && !opts.ignoreErrors {
 		return fmt.Errorf("error installing NVIDIA container runtime hook: %v", err)
 	} else if err != nil {
 		log.Errorf("Ignoring error: %v", fmt.Errorf("error installing NVIDIA container runtime hook: %v", err))
 	}
 
-	nvidiaCTKPath, err := installContainerToolkitCLI(toolkitRoot)
+	nvidiaCTKPath, err := installContainerToolkitCLI(sourceRoot, toolkitRoot)
 	if err != nil && !opts.ignoreErrors {
 		return fmt.Errorf("error installing NVIDIA Container Toolkit CLI: %v", err)
 	} else if err != nil {
 		log.Errorf("Ignoring error: %v", fmt.Errorf("error installing NVIDIA Container Toolkit CLI: %v", err))
 	}
 
-	nvidiaCDIHookPath, err := installContainerCDIHookCLI(toolkitRoot)
+	nvidiaCDIHookPath, err := installContainerCDIHookCLI(sourceRoot, toolkitRoot)
 	if err != nil && !opts.ignoreErrors {
 		return fmt.Errorf("error installing NVIDIA Container CDI Hook CLI: %v", err)
 	} else if err != nil {
@@ -378,7 +379,7 @@ func Install(cli *cli.Context, opts *Options, toolkitRoot string) error {
 // A predefined set of library candidates are considered, with the first one
 // resulting in success being installed to the toolkit folder. The install process
 // resolves the symlink for the library and copies the versioned library itself.
-func installContainerLibraries(toolkitRoot string) error {
+func installContainerLibraries(sourceRoot string, toolkitRoot string) error {
 	log.Infof("Installing NVIDIA container library to '%v'", toolkitRoot)
 
 	libs := []string{
@@ -387,7 +388,7 @@ func installContainerLibraries(toolkitRoot string) error {
 	}
 
 	for _, l := range libs {
-		err := installLibrary(l, toolkitRoot)
+		err := installLibrary(l, sourceRoot, toolkitRoot)
 		if err != nil {
 			return fmt.Errorf("failed to install %s: %v", l, err)
 		}
@@ -397,8 +398,8 @@ func installContainerLibraries(toolkitRoot string) error {
 }
 
 // installLibrary installs the specified library to the toolkit directory.
-func installLibrary(libName string, toolkitRoot string) error {
-	libraryPath, err := findLibrary("", libName)
+func installLibrary(libName string, sourceRoot string, toolkitRoot string) error {
+	libraryPath, err := findLibrary(sourceRoot, libName)
 	if err != nil {
 		return fmt.Errorf("error locating NVIDIA container library: %v", err)
 	}
@@ -524,9 +525,9 @@ func installToolkitConfig(c *cli.Context, toolkitConfigPath string, nvidiaContai
 }
 
 // installContainerToolkitCLI installs the nvidia-ctk CLI executable and wrapper.
-func installContainerToolkitCLI(toolkitDir string) (string, error) {
+func installContainerToolkitCLI(sourceRoot string, toolkitDir string) (string, error) {
 	e := executable{
-		source: "/usr/bin/nvidia-ctk",
+		source: filepath.Join(sourceRoot, "/usr/bin/nvidia-ctk"),
 		target: executableTarget{
 			dotfileName: "nvidia-ctk.real",
 			wrapperName: "nvidia-ctk",
@@ -537,9 +538,9 @@ func installContainerToolkitCLI(toolkitDir string) (string, error) {
 }
 
 // installContainerCDIHookCLI installs the nvidia-cdi-hook CLI executable and wrapper.
-func installContainerCDIHookCLI(toolkitDir string) (string, error) {
+func installContainerCDIHookCLI(sourceRoot string, toolkitDir string) (string, error) {
 	e := executable{
-		source: "/usr/bin/nvidia-cdi-hook",
+		source: filepath.Join(sourceRoot, "/usr/bin/nvidia-cdi-hook"),
 		target: executableTarget{
 			dotfileName: "nvidia-cdi-hook.real",
 			wrapperName: "nvidia-cdi-hook",
@@ -551,15 +552,15 @@ func installContainerCDIHookCLI(toolkitDir string) (string, error) {
 
 // installContainerCLI sets up the NVIDIA container CLI executable, copying the executable
 // and implementing the required wrapper
-func installContainerCLI(toolkitRoot string) (string, error) {
+func installContainerCLI(sourceRoot string, toolkitRoot string) (string, error) {
 	log.Infof("Installing NVIDIA container CLI from '%v'", nvidiaContainerCliSource)
 
 	env := map[string]string{
 		"LD_LIBRARY_PATH": toolkitRoot,
 	}
 
 	e := executable{
-		source: nvidiaContainerCliSource,
+		source: filepath.Join(sourceRoot, nvidiaContainerCliSource),
 		target: executableTarget{
 			dotfileName: "nvidia-container-cli.real",
 			wrapperName: "nvidia-container-cli",
@@ -576,15 +577,15 @@ func installContainerCLI(toolkitRoot string) (string, error) {
 
 // installRuntimeHook sets up the NVIDIA runtime hook, copying the executable
 // and implementing the required wrapper
-func installRuntimeHook(toolkitRoot string, configFilePath string) (string, error) {
+func installRuntimeHook(sourceRoot string, toolkitRoot string, configFilePath string) (string, error) {
 	log.Infof("Installing NVIDIA container runtime hook from '%v'", nvidiaContainerRuntimeHookSource)
 
 	argLines := []string{
 		fmt.Sprintf("-config \"%s\"", configFilePath),
 	}
 
 	e := executable{
-		source: nvidiaContainerRuntimeHookSource,
+		source: filepath.Join(sourceRoot, nvidiaContainerRuntimeHookSource),
 		target: executableTarget{
 			dotfileName: "nvidia-container-runtime-hook.real",
 			wrapperName: "nvidia-container-runtime-hook",
diff --git a/tools/container/toolkit/toolkit_test.go b/tools/container/toolkit/toolkit_test.go
