Retry macOS route command (#1817)

client/internal/routemanager/systemops_darwin.go

@@ -8,7 +8,9 @@ import (
 	"net/netip"
 	"os/exec"
 	"strings"
+	"time"
 
+	"github.com/cenkalti/backoff/v4"
 	log "github.com/sirupsen/logrus"
 
 	"github.com/netbirdio/netbird/client/internal/peer"
@@ -55,11 +57,33 @@ func routeCmd(action string, prefix netip.Prefix, nexthop netip.Addr, intf strin
 		args = append(args, "-interface", intf)
 	}
 
-	out, err := exec.Command("route", args...).CombinedOutput()
-	log.Tracef("route %s: %s", strings.Join(args, " "), out)
-
-	if err != nil {
+	if err := retryRouteCmd(args); err != nil {
 		return fmt.Errorf("failed to %s route for %s: %w", action, prefix, err)
 	}
 	return nil
 }
+
+func retryRouteCmd(args []string) error {
+	operation := func() error {
+		out, err := exec.Command("route", args...).CombinedOutput()
+		log.Tracef("route %s: %s", strings.Join(args, " "), out)
+		// https://github.com/golang/go/issues/45736
+		if err != nil && strings.Contains(string(out), "sysctl: cannot allocate memory") {
+			return err
+		} else if err != nil {
+			return backoff.Permanent(err)
+		}
+		return nil
+	}
+
+	expBackOff := backoff.NewExponentialBackOff()
+	expBackOff.InitialInterval = 50 * time.Millisecond
+	expBackOff.MaxInterval = 500 * time.Millisecond
+	expBackOff.MaxElapsedTime = 1 * time.Second
+
+	err := backoff.Retry(operation, expBackOff)
+	if err != nil {
+		return fmt.Errorf("route cmd retry failed: %w", err)
+	}
+	return nil
+}

client/internal/routemanager/systemops_darwin_test.go

@@ -5,8 +5,10 @@ package routemanager
 import (
 	"fmt"
 	"net"
+	"net/netip"
 	"os/exec"
 	"regexp"
+	"sync"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -29,6 +31,42 @@ func init() {
 	}...)
 }
 
+func TestConcurrentRoutes(t *testing.T) {
+	baseIP := netip.MustParseAddr("192.0.2.0")
+	intf := "lo0"
+
+	var wg sync.WaitGroup
+	for i := 0; i < 1024; i++ {
+		wg.Add(1)
+		go func(ip netip.Addr) {
+			defer wg.Done()
+			prefix := netip.PrefixFrom(ip, 32)
+			if err := addToRouteTable(prefix, netip.Addr{}, intf); err != nil {
+				t.Errorf("Failed to add route for %s: %v", prefix, err)
+			}
+		}(baseIP)
+		baseIP = baseIP.Next()
+	}
+
+	wg.Wait()
+
+	baseIP = netip.MustParseAddr("192.0.2.0")
+
+	for i := 0; i < 1024; i++ {
+		wg.Add(1)
+		go func(ip netip.Addr) {
+			defer wg.Done()
+			prefix := netip.PrefixFrom(ip, 32)
+			if err := removeFromRouteTable(prefix, netip.Addr{}, intf); err != nil {
+				t.Errorf("Failed to remove route for %s: %v", prefix, err)
+			}
+		}(baseIP)
+		baseIP = baseIP.Next()
+	}
+
+	wg.Wait()
+}
+
 func createAndSetupDummyInterface(t *testing.T, intf string, ipAddressCIDR string) string {
 	t.Helper()