fix(module): add idempotency for usb cleanup in virtualization-dra

images/virtualization-dra/internal/plugin/driver.go

@@ -205,6 +205,8 @@ func (d *Driver) UnprepareResourceClaims(ctx context.Context, claims []kubeletpl
 
 func (d *Driver) unprepareResourceClaim(ctx context.Context, claim kubeletplugin.NamespacedObject) error {
 	if err := d.allocator.Unprepare(ctx, claim.UID); err != nil {
+		d.log.Error("error unpreparing devices for claim", slog.Any("error", err), slog.String("uid", string(claim.UID)))
+
 		return fmt.Errorf("error unpreparing devices for claim %v: %w", claim.UID, err)
 	}
 

images/virtualization-dra/internal/usb-gateway/attach_record.go

@@ -118,19 +118,20 @@ func (r *attachRecordManager) Refresh() error {
 		return err
 	}
 
-	// keep only real entries
-	var newEntries []AttachEntry
+	newEntries := make([]AttachEntry, 0, len(record.Entries))
 	for _, e := range record.Entries {
 		if _, ok := ports[e.Rhport]; ok {
 			newEntries = append(newEntries, e)
 		}
 	}
 
-	record.Entries = newEntries
-
-	r.record = record
+	newRecord := attachRecord{Entries: newEntries}
+	if slices.Equal(record.Entries, newRecord.Entries) {
+		r.record = newRecord
+		return nil
+	}
 
-	return nil
+	return r.storeLocked(newRecord)
 }
 
 func (r *attachRecordManager) GetEntries() []AttachEntry {
@@ -163,8 +164,10 @@ func (r *attachRecordManager) AddEntry(e AttachEntry) error {
 	newEntries := slices.Clone(r.record.Entries)
 	newEntries = append(newEntries, e)
 
-	record := attachRecord{Entries: newEntries}
+	return r.storeLocked(attachRecord{Entries: newEntries})
+}
 
+func (r *attachRecordManager) storeLocked(record attachRecord) error {
 	b, err := json.Marshal(record)
 	if err != nil {
 		return err

images/virtualization-dra/internal/usb-gateway/usbgateway.go

@@ -101,7 +101,9 @@ func (c *USBGatewayController) Detach(deviceName string) error {
 	}
 
 	entry := c.findEntry(deviceName)
-	if entry != nil {
+	if entry == nil {
+		log.Info("Device is already detached")
+	} else {
 		log.Info("Detaching USB device")
 		err = c.usbIP.Detach(entry.Rhport)
 		if err != nil {
@@ -115,6 +117,10 @@ func (c *USBGatewayController) Detach(deviceName string) error {
 		return fmt.Errorf("failed to unexport device %s: %w", deviceName, err)
 	}
 
+	if err := c.attachRecordManager.Refresh(); err != nil {
+		return fmt.Errorf("failed to Refresh attach record after detach for device %s: %w", deviceName, err)
+	}
+
 	return nil
 }
 

images/virtualization-dra/internal/usb/store.go

@@ -436,29 +436,34 @@ func (s *AllocationStore) Unprepare(_ context.Context, claimUID types.UID) error
 		return fmt.Errorf("unprepare called before synchronize NRI Hook")
 	}
 
-	usbGatewayEnabled := featuregates.Default().USBGatewayEnabled()
-
-	allocatedDevices := s.resourceClaimAllocations[claimUID]
-	s.log.Info("Unpreparing devices", slog.Any("devices", allocatedDevices), slog.String("claimUID", string(claimUID)))
-
-	for _, device := range allocatedDevices {
-		if usbGatewayEnabled {
-			count := s.usbipAllocatedDevicesCount[device]
-			s.log.Info("Device attached by USBGateway", slog.String("device", device), slog.Int("count", count))
-			if count <= 1 {
-				s.log.Info("Detaching device because has only one consumer", slog.String("device", device))
-				if err := s.usbGateway.Detach(device); err != nil {
-					return fmt.Errorf("failed to detach device %s: %w", device, err)
+	allocatedDevices, exists := s.resourceClaimAllocations[claimUID]
+	if !exists || len(allocatedDevices) == 0 {
+		s.log.Info("Claim has no tracked allocations, skipping device cleanup", slog.String("claimUID", string(claimUID)))
+	} else {
+		usbGatewayEnabled := featuregates.Default().USBGatewayEnabled()
+
+		s.log.Info("Unpreparing devices", slog.Any("devices", allocatedDevices), slog.String("claimUID", string(claimUID)))
+
+		for _, device := range allocatedDevices {
+			if usbGatewayEnabled {
+				count, hasCount := s.usbipAllocatedDevicesCount[device]
+				s.log.Info("Device attached by USBGateway", slog.String("device", device), slog.Int("count", count))
+				switch {
+				case !hasCount || count <= 1:
+					s.log.Info("Device has no tracked consumers, attempting detach cleanup", slog.String("device", device), slog.Int("count", count))
+					if err := s.usbGateway.Detach(device); err != nil {
+						return fmt.Errorf("failed to detach device %s: %w", device, err)
+					}
+					delete(s.usbipAllocatedDevicesCount, device)
+				default:
+					s.log.Info("Decrementing device consumer count", slog.String("device", device), slog.Int("newCount", count-1))
+					s.usbipAllocatedDevicesCount[device]--
 				}
-				delete(s.usbipAllocatedDevicesCount, device)
-			} else {
-				s.log.Info("Decrementing device consumer count", slog.String("device", device), slog.Int("newCount", count-1))
-				s.usbipAllocatedDevicesCount[device]--
 			}
-		}
 
-		s.log.Info("Deleting device from allocated devices", slog.String("device", device))
-		s.allocatedDevices.Delete(device)
+			s.log.Info("Deleting device from allocated devices", slog.String("device", device))
+			s.allocatedDevices.Delete(device)
+		}
 	}
 
 	s.log.Info("Deleting CDI claim spec file", slog.String("claimUID", string(claimUID)))

images/virtualization-dra/pkg/usbip/exporter.go

@@ -98,11 +98,12 @@ func (e *usbExporter) Unexport(host, busID string, port int) error {
 		return fmt.Errorf("unsupported USBIP version: %d", unExportReply.Version)
 	}
 
-	if unExportReply.Status != protocol.OpStatusOk {
+	switch unExportReply.Status {
+	case protocol.OpStatusOk, protocol.OpStatusNoDev:
+		return nil
+	default:
 		return fmt.Errorf("reply failed: %s", unExportReply.Status.String())
 	}
-
-	return nil
 }
 
 func (e *usbExporter) usbipNetTCPConnect(host string, port int) (*net.TCPConn, error) {