p2p: retry failed connections slightly more aggressively (backport #8010) (#8012)

internal/p2p/peermanager.go

@@ -1038,13 +1038,15 @@ func (m *PeerManager) retryDelay(failures uint32, persistent bool) time.Duration
 		maxDelay = m.options.MaxRetryTimePersistent
 	}
 
-	delay := m.options.MinRetryTime * time.Duration(math.Pow(2, float64(failures-1)))
-	if maxDelay > 0 && delay > maxDelay {
-		delay = maxDelay
-	}
+	delay := m.options.MinRetryTime * time.Duration(failures)
 	if m.options.RetryTimeJitter > 0 {
 		delay += time.Duration(m.rand.Int63n(int64(m.options.RetryTimeJitter)))
 	}
+
+	if maxDelay > 0 && delay > maxDelay {
+		delay = maxDelay
+	}
+
 	return delay
 }
 

internal/p2p/peermanager_test.go

@@ -3,6 +3,7 @@ package p2p_test
 import (
 	"context"
 	"errors"
+	"fmt"
 	"strings"
 	"testing"
 	"time"
@@ -314,13 +315,14 @@ func TestPeerManager_DialNext_Retry(t *testing.T) {
 	ctx, cancel := context.WithTimeout(ctx, 5*time.Second)
 	defer cancel()
 
-	for i := 0; i <= 5; i++ {
+	for i := 0; i <= 6; i++ {
 		start := time.Now()
 		dial, err := peerManager.DialNext(ctx)
 		require.NoError(t, err)
 		require.Equal(t, a, dial)
 		elapsed := time.Since(start).Round(time.Millisecond)
 
+		fmt.Println(elapsed, options.MinRetryTime)
 		switch i {
 		case 0:
 			require.LessOrEqual(t, elapsed, options.MinRetryTime)
@@ -329,12 +331,11 @@ func TestPeerManager_DialNext_Retry(t *testing.T) {
 		case 2:
 			require.GreaterOrEqual(t, elapsed, 2*options.MinRetryTime)
 		case 3:
+			require.GreaterOrEqual(t, elapsed, 3*options.MinRetryTime)
+		case 4, 5, 6:
 			require.GreaterOrEqual(t, elapsed, 4*options.MinRetryTime)
-		case 4, 5:
-			require.GreaterOrEqual(t, elapsed, options.MaxRetryTime)
-			require.LessOrEqual(t, elapsed, 8*options.MinRetryTime)
 		default:
-			require.Fail(t, "unexpected retry")
+			t.Fatal("unexpected retry")
 		}
 
 		require.NoError(t, peerManager.DialFailed(a))

node/setup.go

@@ -461,10 +461,10 @@ func createPeerManager(
 		MaxConnected:           maxConns,
 		MaxConnectedUpgrade:    4,
 		MaxPeers:               1000,
-		MinRetryTime:           100 * time.Millisecond,
-		MaxRetryTime:           8 * time.Hour,
+		MinRetryTime:           250 * time.Millisecond,
+		MaxRetryTime:           30 * time.Minute,
 		MaxRetryTimePersistent: 5 * time.Minute,
-		RetryTimeJitter:        3 * time.Second,
+		RetryTimeJitter:        5 * time.Second,
 		PrivatePeers:           privatePeerIDs,
 	}