Fixing pcidevice device plugin stop deadlock

Signed-off-by: Webber Huang <webber.huang@suse.com>

Author: WebberHuang1118

pkg/controller/nodes/node_controller.go

@@ -7,8 +7,6 @@ import (
 	"reflect"
 	"time"
 
-	"github.com/harvester/pcidevices/pkg/controller/gpudevice"
-
 	ctlnetworkv1beta1 "github.com/harvester/harvester-network-controller/pkg/generated/controllers/network.harvesterhci.io/v1beta1"
 	"github.com/jaypipes/ghw"
 	ctlcorev1 "github.com/rancher/wrangler/pkg/generated/controllers/core/v1"
@@ -17,6 +15,7 @@ import (
 	"k8s.io/apimachinery/pkg/labels"
 
 	"github.com/harvester/pcidevices/pkg/apis/devices.harvesterhci.io/v1beta1"
+	"github.com/harvester/pcidevices/pkg/controller/gpudevice"
 	"github.com/harvester/pcidevices/pkg/controller/pcidevice"
 	"github.com/harvester/pcidevices/pkg/controller/sriovdevice"
 	ctl "github.com/harvester/pcidevices/pkg/generated/controllers/devices.harvesterhci.io/v1beta1"

pkg/deviceplugins/device_manager.go

@@ -45,6 +45,7 @@ const (
 	pciBasePath       = "/sys/bus/pci/devices"
 	connectionTimeout = 120 * time.Second // Google gRPC default timeout
 	PCIResourcePrefix = "PCI_RESOURCE"
+	tickerTimeout     = 30 * time.Second
 )
 
 type PCIDevice struct {
@@ -293,7 +294,7 @@ func (dp *PCIDevicePlugin) Allocate(_ context.Context, r *pluginapi.AllocateRequ
 
 func (dp *PCIDevicePlugin) healthCheck() error {
 	logger := log.DefaultLogger()
-	monitoredDevices := make(map[string]string)
+	monitoredDevices := map[string]string{}
 	watcher, err := fsnotify.NewWatcher()
 	if err != nil {
 		return fmt.Errorf("failed to creating a fsnotify watcher: %v", err)
@@ -352,6 +353,8 @@ func (dp *PCIDevicePlugin) healthCheck() error {
 		select {
 		case <-dp.stop:
 			return nil
+		case <-dp.done:
+			return nil
 		case err := <-watcher.Errors:
 			logger.Reason(err).Errorf("error watching devices and device plugin directory")
 		case event := <-watcher.Events:
@@ -389,14 +392,12 @@ func (dp *PCIDevicePlugin) GetDeviceName() string {
 
 // Stop stops the gRPC server
 func (dp *PCIDevicePlugin) stopDevicePlugin() error {
-	defer func() {
-		if !IsChanClosed(dp.done) {
-			close(dp.done)
-		}
-	}()
+	if !IsChanClosed(dp.done) {
+		close(dp.done)
+	}
 
 	// Give the device plugin one second to properly deregister
-	ticker := time.NewTicker(1 * time.Second)
+	ticker := time.NewTicker(tickerTimeout)
 	defer ticker.Stop()
 	select {
 	case <-dp.deregistered:

pkg/deviceplugins/deviceplugin.go

@@ -92,5 +92,11 @@ func (dp *PCIDevicePlugin) RemoveDevice(pd *v1beta1.PCIDevice, pdc *v1beta1.PCID
 		logrus.Infof("Removing %s from device plugin", resourceName)
 		dp.MarkPCIDeviceAsUnhealthy(pdc.Spec.Address)
 	}
+
+	for i, dev := range dp.devs {
+		if dev.ID == pdc.Spec.Address {
+			dp.devs[i].Health = pluginapi.Unhealthy
+		}
+	}
 	return nil
 }