Merge pull request #65 from telekom/fix/frr-systemd-reload

Wait for FRR reload to complete, restart if failed

pkg/frr/manager.go

@@ -92,10 +92,32 @@ func (*Manager) ReloadFRR() error {
 	}
 	defer con.Close()
 
-	_, err = con.ReloadUnitContext(context.Background(), frrUnit, "fail", nil)
-	if err != nil {
+	jobChan := make(chan string)
+	if _, err = con.ReloadUnitContext(context.Background(), frrUnit, "fail", jobChan); err != nil {
 		return fmt.Errorf("error reloading %s context: %w", frrUnit, err)
 	}
+	reloadStatus := <-jobChan
+	if reloadStatus != "done" {
+		return fmt.Errorf("error reloading %s, job status is %s", frrUnit, reloadStatus)
+	}
+	return nil
+}
+
+func (*Manager) RestartFRR() error {
+	con, err := dbus.NewSystemConnectionContext(context.Background())
+	if err != nil {
+		return fmt.Errorf("error creating nee D-Bus connection: %w", err)
+	}
+	defer con.Close()
+
+	jobChan := make(chan string)
+	if _, err = con.RestartUnitContext(context.Background(), frrUnit, "fail", jobChan); err != nil {
+		return fmt.Errorf("error restarting %s context: %w", frrUnit, err)
+	}
+	restartStatus := <-jobChan
+	if restartStatus != "done" {
+		return fmt.Errorf("error restarting %s, job status is %s", restartStatus, restartStatus)
+	}
 	return nil
 }
 

pkg/reconciler/layer3.go

@@ -65,11 +65,17 @@ func (r *reconcile) reconcileLayer3(l3vnis []networkv1alpha1.VRFRouteConfigurati
 		r.Logger.Info("trying to reload FRR config because it changed")
 		err = r.frrManager.ReloadFRR()
 		if err != nil {
-			r.dirtyFRRConfig = true
-			r.Logger.Error(err, "error reloading FRR systemd unit")
-			return fmt.Errorf("error reloading FRR systemd unit: %w", err)
+			r.Logger.Error(err, "error reloading FRR systemd unit, trying restart")
+
+			err = r.frrManager.RestartFRR()
+			if err != nil {
+				r.dirtyFRRConfig = true
+				r.Logger.Error(err, "error restarting FRR systemd unit")
+				return fmt.Errorf("error reloading / restarting FRR systemd unit: %w", err)
+			}
 		}
 		r.dirtyFRRConfig = false
+		r.Logger.Info("reloaded FRR config")
 	}
 
 	// Make sure that all created netlink VRFs are up after FRR reload