Clean Systemd files on exit

cmd/sriov-network-config-daemon/start.go

@@ -21,7 +21,9 @@ import (
 	"net"
 	"net/url"
 	"os"
+	"os/signal"
 	"strings"
+	"syscall"
 	"time"
 
 	"github.com/spf13/cobra"
@@ -134,7 +136,6 @@ func runStartCmd(cmd *cobra.Command, args []string) error {
 
 	// This channel is used to ensure all spawned goroutines exit when we exit.
 	stopCh := make(chan struct{})
-	defer close(stopCh)
 
 	// This channel is used to signal Run() something failed and to jump ship.
 	// It's purely a chan<- in the Daemon struct for goroutines to write to, and
@@ -286,6 +287,7 @@ func runStartCmd(cmd *cobra.Command, args []string) error {
 	err = kClient.Get(context.Background(), types.NamespacedName{Namespace: vars.Namespace, Name: consts.DefaultConfigName}, defaultConfig)
 	if err != nil {
 		log.Log.Error(err, "Failed to get default SriovOperatorConfig object")
+		close(stopCh)
 		return err
 	}
 	featureGates := featuregate.New()
@@ -294,25 +296,45 @@ func runStartCmd(cmd *cobra.Command, args []string) error {
 	log.Log.Info("Enabled featureGates", "featureGates", featureGates.String())
 
 	setupLog.V(0).Info("Starting SriovNetworkConfigDaemon")
-	err = daemon.New(
-		kClient,
-		snclient,
-		kubeclient,
-		hostHelpers,
-		platformHelper,
-		exitCh,
-		stopCh,
-		syncCh,
-		refreshCh,
-		eventRecorder,
-		featureGates,
-		startOpts.disabledPlugins,
-	).Run(stopCh, exitCh)
-	if err != nil {
-		setupLog.Error(err, "failed to run daemon")
+
+	// create a signal channel to catch interrupts and gracefully shutdown the daemon
+	sigc := make(chan os.Signal, 1)
+	signal.Notify(sigc, os.Interrupt)
+	signal.Notify(sigc, syscall.SIGTERM)
+
+	errChan := make(chan error)
+	defer close(errChan)
+	go func() {
+		errChan <- daemon.New(
+			kClient,
+			snclient,
+			kubeclient,
+			hostHelpers,
+			platformHelper,
+			exitCh,
+			stopCh,
+			syncCh,
+			refreshCh,
+			eventRecorder,
+			featureGates,
+			startOpts.disabledPlugins,
+		).Run(stopCh, exitCh)
+	}()
+
+	select {
+	case err := <-errChan:
+		// daemon has exited, close the stop channel and return the error
+		close(stopCh)
+		return err
+	case <-sigc:
+		// signal received, close the stop channel and wait for the daemon to exit
+		close(stopCh)
+		if err := <-errChan; err != nil {
+			return err
+		}
 	}
 	setupLog.V(0).Info("Shutting down SriovNetworkConfigDaemon")
-	return err
+	return nil
 }
 
 // updateDialer instruments a restconfig with a dial. the returned function allows forcefully closing all active connections.

pkg/daemon/daemon.go

@@ -77,7 +77,7 @@ type Daemon struct {
 
 	refreshCh chan<- Message
 
-	mu *sync.Mutex
+	mu sync.Mutex
 
 	disableDrain bool
 
@@ -120,7 +120,6 @@ func New(
 		eventRecorder:   er,
 		featureGate:     featureGates,
 		disabledPlugins: disabledPlugins,
-		mu:              &sync.Mutex{},
 	}
 }
 
@@ -207,6 +206,20 @@ func (dn *Daemon) Run(stopCh <-chan struct{}, exitCh <-chan error) error {
 	for {
 		select {
 		case <-stopCh:
+			// clean files from host if we are running in systemd mode and the node
+			// is not required to be rebooted
+			dn.mu.Lock()
+			rebootrequired := utils.ObjectHasAnnotation(dn.desiredNodeState,
+				consts.NodeStateDrainAnnotation, consts.RebootRequired)
+			dn.mu.Unlock()
+
+			if vars.UsingSystemdMode && !rebootrequired {
+				err := systemd.CleanSriovFilesFromHost(vars.ClusterType == consts.ClusterTypeOpenshift)
+				if err != nil {
+					log.Log.Error(err, "failed to remove all the systemd sriov files")
+					return err
+				}
+			}
 			log.Log.V(0).Info("Run(): stop daemon")
 			return nil
 		case err, more := <-exitCh:
@@ -316,6 +329,8 @@ func (dn *Daemon) operatorConfigChangeHandler(old, new interface{}) {
 }
 
 func (dn *Daemon) nodeStateSyncHandler() error {
+	dn.mu.Lock()
+	defer dn.mu.Unlock()
 	var err error
 	// Get the latest NodeState
 	var sriovResult = &systemd.SriovResult{SyncStatus: consts.SyncStatusSucceeded, LastSyncError: ""}
@@ -679,8 +694,6 @@ func (dn *Daemon) handleDrain(reqReboot bool) (bool, error) {
 }
 
 func (dn *Daemon) restartDevicePluginPod() error {
-	dn.mu.Lock()
-	defer dn.mu.Unlock()
 	log.Log.V(2).Info("restartDevicePluginPod(): try to restart device plugin pod")
 
 	pods, err := dn.kubeClient.CoreV1().Pods(vars.Namespace).List(context.Background(), metav1.ListOptions{