Honor fail-on-init-error when no resources are found

As implemented GFD will not fail if no resources are detected -- even if fail-on-init-error is set. This change ensures that fail-on-init-error is honored if no resources are detected. Signed-off-by: Evan Lezar <[email protected]>
NVIDIA · Oct 31, 2024 · 6413d86 · 6413d86
1 parent 78c5937
commit 6413d86
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 8 deletions.
diff --git a/cmd/gpu-feature-discovery/main.go b/cmd/gpu-feature-discovery/main.go
@@ -155,7 +155,10 @@ func start(c *cli.Context, cfg *Config) error {
 		}
 		klog.Infof("\nRunning with config:\n%v", string(configJSON))
 
-		manager := resource.NewManager(config)
+		manager, err := resource.NewManager(config)
+		if err != nil {
+			return err
+		}
 		vgpul := vgpu.NewVGPULib(vgpu.NewNvidiaPCILib())
 
 		var clientSets flags.ClientSets

diff --git a/internal/resource/factory.go b/internal/resource/factory.go
@@ -17,15 +17,26 @@
 package resource
 
 import (
+	"fmt"
+
 	"github.com/NVIDIA/go-nvlib/pkg/nvlib/info"
 	"k8s.io/klog/v2"
 
 	spec "github.com/NVIDIA/k8s-device-plugin/api/config/v1"
 )
 
 // NewManager is a factory method that creates a resource Manager based on the specified config.
-func NewManager(config *spec.Config) Manager {
-	return WithConfig(getManager(), config)
+func NewManager(config *spec.Config) (Manager, error) {
+	manager, err := getManager()
+	if err != nil {
+		if *config.Flags.FailOnInitError {
+			return nil, err
+		}
+		klog.ErrorS(err, "using empty manager")
+		return NewNullManager(), nil
+	}
+
+	return WithConfig(manager, config), nil
 }
 
 // WithConfig modifies a manager depending on the specified config.
@@ -39,7 +50,7 @@ func WithConfig(manager Manager, config *spec.Config) Manager {
 }
 
 // getManager returns the resource manager depending on the system configuration.
-func getManager() Manager {
+func getManager() (Manager, error) {
 	// logWithReason logs the output of the has* / is* checks from the info.Interface
 	logWithReason := func(f func() (bool, string), tag string) bool {
 		is, reason := f()
@@ -63,12 +74,11 @@ func getManager() Manager {
 
 	if hasNVML {
 		klog.Info("Using NVML manager")
-		return NewNVMLManager()
+		return NewNVMLManager(), nil
 	} else if isTegra {
 		klog.Info("Using CUDA manager")
-		return NewCudaManager()
+		return NewCudaManager(), nil
 	}
 
-	klog.Warning("No valid resources detected; using empty manager.")
-	return NewNullManager()
+	return nil, fmt.Errorf("no valid resource detected")
 }