From 61e4cfd1b7600a2b58d05fe592d6bbb56eb997a9 Mon Sep 17 00:00:00 2001
From: Dmitry Shulyak <yashulyak@gmail.com>
Date: Thu, 21 Sep 2023 05:58:55 +0000
Subject: [PATCH] sync: parametrize out of sync threshold and set it to 3h for
 mainnet (#5040)

related: https://github.com/spacemeshos/go-spacemesh/issues/5036

in future we should drop it completely, and use only connectivity information to decide if node should stop participating
in consensus. there should be no risk of interrupting consensus, because of any unexpected failures in sync process.
---
 config/mainnet.go     | 11 ++++++-----
 syncer/syncer.go      | 38 ++++++++++++++++++--------------------
 syncer/syncer_test.go | 15 +++++++++------
 3 files changed, 33 insertions(+), 31 deletions(-)

diff --git a/config/mainnet.go b/config/mainnet.go
index 9b28038c6e..b8472f8a76 100644
--- a/config/mainnet.go
+++ b/config/mainnet.go
@@ -136,11 +136,12 @@ func MainnetConfig() Config {
 		FETCH:    fetch.DefaultConfig(),
 		LOGGING:  logging,
 		Sync: syncer.Config{
-			Interval:         time.Minute,
-			EpochEndFraction: 0.8,
-			MaxStaleDuration: time.Hour,
-			Standalone:       false,
-			GossipDuration:   50 * time.Second,
+			Interval:                 time.Minute,
+			EpochEndFraction:         0.8,
+			MaxStaleDuration:         time.Hour,
+			Standalone:               false,
+			GossipDuration:           50 * time.Second,
+			OutOfSyncThresholdLayers: 36, // 3h
 		},
 		Recovery: checkpoint.DefaultConfig(),
 		Cache:    datastore.DefaultConfig(),
diff --git a/syncer/syncer.go b/syncer/syncer.go
index c840e69db9..faab996b59 100644
--- a/syncer/syncer.go
+++ b/syncer/syncer.go
@@ -21,31 +21,29 @@ import (
 
 // Config is the config params for syncer.
 type Config struct {
-	Interval         time.Duration
-	EpochEndFraction float64
-	HareDelayLayers  uint32
-	SyncCertDistance uint32
-	MaxStaleDuration time.Duration
-	Standalone       bool
-	GossipDuration   time.Duration
+	Interval                 time.Duration
+	EpochEndFraction         float64
+	HareDelayLayers          uint32
+	SyncCertDistance         uint32
+	MaxStaleDuration         time.Duration
+	Standalone               bool
+	GossipDuration           time.Duration
+	OutOfSyncThresholdLayers uint32 `mapstructure:"out-of-sync-threshold"`
 }
 
 // DefaultConfig for the syncer.
 func DefaultConfig() Config {
 	return Config{
-		Interval:         10 * time.Second,
-		EpochEndFraction: 0.8,
-		HareDelayLayers:  10,
-		SyncCertDistance: 10,
-		MaxStaleDuration: time.Second,
-		GossipDuration:   15 * time.Second,
+		Interval:                 10 * time.Second,
+		EpochEndFraction:         0.8,
+		HareDelayLayers:          10,
+		SyncCertDistance:         10,
+		MaxStaleDuration:         time.Second,
+		GossipDuration:           15 * time.Second,
+		OutOfSyncThresholdLayers: 3,
 	}
 }
 
-const (
-	outOfSyncThreshold uint32 = 3 // see notSynced
-)
-
 type syncState uint32
 
 const (
@@ -452,7 +450,7 @@ func (s *Syncer) syncAtx(ctx context.Context) error {
 	return nil
 }
 
-func isTooFarBehind(ctx context.Context, logger log.Log, current, lastSynced types.LayerID) bool {
+func isTooFarBehind(ctx context.Context, logger log.Log, current, lastSynced types.LayerID, outOfSyncThreshold uint32) bool {
 	if current.After(lastSynced) && current.Difference(lastSynced) >= outOfSyncThreshold {
 		logger.WithContext(ctx).With().Info("node is too far behind",
 			log.Stringer("current", current),
@@ -472,7 +470,7 @@ func (s *Syncer) setStateBeforeSync(ctx context.Context) {
 		}
 		return
 	}
-	if isTooFarBehind(ctx, s.logger, current, s.getLastSyncedLayer()) {
+	if isTooFarBehind(ctx, s.logger, current, s.getLastSyncedLayer(), s.cfg.OutOfSyncThresholdLayers) {
 		s.setSyncState(ctx, notSynced)
 	}
 }
@@ -492,7 +490,7 @@ func (s *Syncer) setStateAfterSync(ctx context.Context, success bool) {
 	// network outage.
 	switch currSyncState {
 	case synced:
-		if !success && isTooFarBehind(ctx, s.logger, current, s.getLastSyncedLayer()) {
+		if !success && isTooFarBehind(ctx, s.logger, current, s.getLastSyncedLayer(), s.cfg.OutOfSyncThresholdLayers) {
 			s.setSyncState(ctx, notSynced)
 		}
 	case gossipSync:
diff --git a/syncer/syncer_test.go b/syncer/syncer_test.go
index bd2d36c81a..1cdba9772b 100644
--- a/syncer/syncer_test.go
+++ b/syncer/syncer_test.go
@@ -28,6 +28,8 @@ import (
 const (
 	layersPerEpoch = 3
 	never          = time.Second * 60 * 24
+
+	outOfSyncThreshold = 3
 )
 
 func TestMain(m *testing.M) {
@@ -93,11 +95,12 @@ func newTestSyncer(t *testing.T, interval time.Duration) *testSyncer {
 	require.NoError(t, err)
 
 	cfg := Config{
-		Interval:         interval,
-		GossipDuration:   5 * time.Millisecond,
-		EpochEndFraction: 0.66,
-		SyncCertDistance: 4,
-		HareDelayLayers:  5,
+		Interval:                 interval,
+		GossipDuration:           5 * time.Millisecond,
+		EpochEndFraction:         0.66,
+		SyncCertDistance:         4,
+		HareDelayLayers:          5,
+		OutOfSyncThresholdLayers: outOfSyncThreshold,
 	}
 	ts.syncer = NewSyncer(ts.cdb, ts.mTicker, ts.mBeacon, ts.msh, nil, nil, ts.mLyrPatrol, ts.mCertHdr,
 		WithConfig(cfg),
@@ -535,7 +538,7 @@ func TestNetworkHasNoData(t *testing.T) {
 		require.True(t, ts.syncer.IsSynced(context.Background()))
 	}
 	// the network hasn't received any data
-	require.Greater(t, ts.syncer.ticker.CurrentLayer()-ts.msh.LatestLayer(), outOfSyncThreshold)
+	require.Greater(t, int(ts.syncer.ticker.CurrentLayer()-ts.msh.LatestLayer()), outOfSyncThreshold)
 }
 
 // test the case where the node was originally synced, and somehow gets out of sync, but