Changed FRR configuration behavior

Signed-off-by: Patryk Strusiewicz-Surmacki <[email protected]>

pkg/frr/configure.go

@@ -43,7 +43,7 @@ func (m *Manager) Configure(in Configuration, nm *nl.Manager, nwopCfg *config.Co
 		for j := range in.VRFs[i].Import {
 			for k := range in.VRFs[i].Import[j].Items {
 				if in.VRFs[i].Import[j].Items[k].Action != "deny" {
-					return false, fmt.Errorf("only deny rules are allowed in import prefix-lists of mgmt VRFs")
+					return false, &ConfigurationError{fmt.Errorf("only deny rules are allowed in import prefix-lists of mgmt VRFs")}
 				}
 				// Swap deny to permit, this will be a prefix-list called from a deny route-map
 				in.VRFs[i].Import[j].Items[k].Action = "permit"
@@ -58,7 +58,7 @@ func (m *Manager) Configure(in Configuration, nm *nl.Manager, nwopCfg *config.Co
 
 	currentConfig, err := os.ReadFile(m.ConfigPath)
 	if err != nil {
-		return false, fmt.Errorf("error reading configuration file: %w", err)
+		return false, &ConfigurationError{fmt.Errorf("error reading configuration file: %w", err)}
 	}
 
 	targetConfig, err := render(m.configTemplate, frrConfig)
@@ -72,7 +72,7 @@ func (m *Manager) Configure(in Configuration, nm *nl.Manager, nwopCfg *config.Co
 	if !bytes.Equal(currentConfig, targetConfig) {
 		err = os.WriteFile(m.ConfigPath, targetConfig, frrPermissions)
 		if err != nil {
-			return false, fmt.Errorf("error writing configuration file: %w", err)
+			return false, &ConfigurationError{fmt.Errorf("error writing configuration file: %w", err)}
 		}
 
 		return true, nil
@@ -187,3 +187,11 @@ func applyCfgReplacements(frrConfig []byte, replacements []config.Replacement) [
 	}
 	return frrConfig
 }
+
+type ConfigurationError struct {
+	err error
+}
+
+func (r *ConfigurationError) Error() string {
+	return fmt.Sprintf("FRR configuration error: %s", r.err.Error())
+}

pkg/reconciler/layer3.go

@@ -2,6 +2,7 @@ package reconciler
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net"
 	"sort"
@@ -65,6 +66,15 @@ func (r *reconcile) reconcileLayer3(l3vnis []networkv1alpha1.VRFRouteConfigurati
 		return allConfigs[i].VNI < allConfigs[j].VNI
 	})
 
+	// Create FRR configuration and reload it
+	err = r.configureFRR(allConfigs)
+	if err != nil {
+		if !errors.Is(err, &frr.ConfigurationError{}) {
+			return err
+		}
+		r.Logger.Error(err, "failed to configure FRR")
+	}
+
 	created, deletedVRF, err := r.reconcileL3Netlink(l3Configs)
 	if err != nil {
 		r.Logger.Error(err, "error reconciling Netlink")
@@ -75,54 +85,40 @@ func (r *reconcile) reconcileLayer3(l3vnis []networkv1alpha1.VRFRouteConfigurati
 	if err != nil {
 		return err
 	}
-	reloadTwice := deletedVRF || deletedTaas
 
-	// We wait here for two seconds to let FRR settle after updating netlink devices
-	time.Sleep(defaultSleep)
-
-	err = r.configureFRR(allConfigs, reloadTwice)
-	if err != nil {
-		return err
+	// When a BGP VRF is deleted there is a leftover running configuration after reload
+	// A second reload fixes this.
+	if deletedVRF || deletedTaas {
+		if err := r.reloadFRR(); err != nil {
+			return fmt.Errorf("failed to reload FRR: %w", err)
+		}
 	}
 
-	// Make sure that all created netlink VRFs are up after FRR reload
+	// We wait here for two seconds to let FRR settle after updating netlink devices
 	time.Sleep(defaultSleep)
+
 	for _, info := range created {
 		if err := r.netlinkManager.UpL3(info); err != nil {
-			r.Logger.Error(err, "error setting L3 to state UP")
-			return fmt.Errorf("error setting L3 to state UP: %w", err)
+			r.Logger.Error(err, "error setting L3 to state UP", "interface", info)
 		}
 	}
 	return nil
 }
 
-func (r *reconcile) configureFRR(vrfConfigs []frr.VRFConfiguration, reloadTwice bool) error {
+func (r *reconcile) configureFRR(vrfConfigs []frr.VRFConfiguration) error {
 	changed, err := r.frrManager.Configure(frr.Configuration{
 		VRFs: vrfConfigs,
 		ASN:  r.config.ServerASN,
 	}, r.netlinkManager, r.config)
 	if err != nil {
-		r.Logger.Error(err, "error updating FRR configuration")
 		return fmt.Errorf("error updating FRR configuration: %w", err)
 	}
 
-	if changed || r.dirtyFRRConfig {
+	if changed {
 		err := r.reloadFRR()
 		if err != nil {
-			r.dirtyFRRConfig = true
 			return err
 		}
-
-		// When a BGP VRF is deleted there is a leftover running configuration after reload
-		// A second reload fixes this.
-		if reloadTwice {
-			err := r.reloadFRR()
-			if err != nil {
-				r.dirtyFRRConfig = true
-				return err
-			}
-		}
-		r.dirtyFRRConfig = false
 	}
 	return nil
 }
@@ -249,13 +245,17 @@ func (r *reconcile) reconcileL3Netlink(vrfConfigs []frr.VRFConfiguration) ([]nl.
 	}
 
 	// Check for VRFs that are configured on the host but no longer in Kubernetes
+	preexisting := []nl.VRFInformation{}
 	toDelete := []nl.VRFInformation{}
 	for i := range existing {
 		stillExists := false
 		for j := range vrfConfigs {
 			if vrfConfigs[j].Name == existing[i].Name && vrfConfigs[j].VNI == existing[i].VNI {
 				stillExists = true
 				existing[i].MTU = vrfConfigs[j].MTU
+				if !existing[i].MarkForDelete {
+					preexisting = append(preexisting, existing[i])
+				}
 				break
 			}
 		}
@@ -266,6 +266,13 @@ func (r *reconcile) reconcileL3Netlink(vrfConfigs []frr.VRFConfiguration) ([]nl.
 		}
 	}
 
+	// Make sure that all previously configured L3 interfaces are up
+	for _, info := range preexisting {
+		if err := r.netlinkManager.UpL3(info); err != nil {
+			r.Logger.Error(err, "failed to set L3 up", "interface", info.Name)
+		}
+	}
+
 	// Check for VRFs that are in Kubernetes but not yet configured on the host
 	toCreate := prepareVRFsToCreate(vrfConfigs, existing)
 

pkg/reconciler/reconciler.go

@@ -28,8 +28,6 @@ type Reconciler struct {
 	healthChecker  *healthcheck.HealthChecker
 
 	debouncer *debounce.Debouncer
-
-	dirtyFRRConfig bool
 }
 
 type reconcile struct {