Adding env var to confirm CDI is enabled

Signed-off-by: Vishesh Tanksale <vtanksale@nvidia.com>

Author: visheshtanksale

cmd/compute-domain-daemon/main.go

@@ -204,6 +204,13 @@ func newApp() *cli.App {
 
 // Run invokes the IMEX daemon and manages its lifecycle.
 func run(ctx context.Context, cancel context.CancelFunc, flags *Flags) error {
+	// Verify that CDI container edits were applied by the container runtime.
+	// If the env var is not set to "true", CDI is likely disabled and the daemon
+	// cannot function correctly (e.g. the /imexd mount will be missing).
+	if os.Getenv("NVIDIA_CDI_EDITS_APPLIED") != "true" {
+		return fmt.Errorf("CDI container edits did not apply -- is CDI enabled in your container runtime?")
+	}
+
 	common.StartDebugSignalHandlers()
 
 	// Validate feature gate dependencies

cmd/compute-domain-kubelet-plugin/computedomain.go

@@ -162,6 +162,9 @@ func (s *ComputeDomainDaemonSettings) GetCDIContainerEditsCommon(ctx context.Con
 	edits := &cdiapi.ContainerEdits{
 		ContainerEdits: &cdispec.ContainerEdits{
 			Env: []string{
+				// This is a value that the CD daemon checks at startup to verify that CDI edits were applied by the container runtime.
+				// If the value is not present, CDI is likely disabled and the daemon cannot function correctly (e.g. the /imexd mount will be missing).
+				"NVIDIA_CDI_EDITS_APPLIED=true",
 				fmt.Sprintf("CLIQUE_ID=%s", s.manager.cliqueID),
 				fmt.Sprintf("COMPUTE_DOMAIN_UUID=%s", cd.UID),
 				fmt.Sprintf("COMPUTE_DOMAIN_NAME=%s", cd.Name),

internal/common/util.go

@@ -29,6 +29,15 @@ import (
 
 const dumpPath = "/tmp/goroutine-stacks.dump"
 
+// CDIEditsAppliedEnvKey and CDIEditsAppliedEnvValue is a key/value
+// pair injected via CDI container edits into the CD daemon container. The CD
+// daemon checks for this key/value pair at startup to verify that CDI edits were
+// applied by the container runtime. If not present, CDI is likely disabled.
+const (
+	CDIEditsAppliedEnvKey   = "NVIDIA_CDI_EDITS_APPLIED"
+	CDIEditsAppliedEnvValue = "true"
+)
+
 // Set up SIGUSR2 handler: if triggered, acquire stack traces for all goroutines
 // in this process. Dump to file, and fall back to emitting to stderr if file
 // output didn't work.