sync: parametrize out of sync threshold and set it to 3h for mainnet

fix tests
spacemeshos · Sep 20, 2023 · 4c69d42 · 4c69d42
1 parent 2b1c6d9
commit 4c69d42
Show file tree

Hide file tree

Showing 3 changed files with 37 additions and 35 deletions.
diff --git a/config/mainnet.go b/config/mainnet.go
@@ -133,12 +133,13 @@ func MainnetConfig() Config {
 		FETCH:    fetch.DefaultConfig(),
 		LOGGING:  defaultLoggingConfig(),
 		Sync: syncer.Config{
-			Interval:         time.Minute,
-			EpochEndFraction: 0.8,
-			MaxStaleDuration: time.Hour,
-			UseNewProtocol:   true,
-			Standalone:       false,
-			GossipDuration:   50 * time.Second,
+			Interval:           time.Minute,
+			EpochEndFraction:   0.8,
+			MaxStaleDuration:   time.Hour,
+			UseNewProtocol:     true,
+			Standalone:         false,
+			GossipDuration:     50 * time.Second,
+			OutOfSyncThreshold: 36,
 		},
 		Recovery: checkpoint.DefaultConfig(),
 		Cache:    datastore.DefaultConfig(),

diff --git a/syncer/syncer.go b/syncer/syncer.go
@@ -21,33 +21,31 @@ 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
-	UseNewProtocol   bool `mapstructure:"use-new-opn"`
-	GossipDuration   time.Duration
+	Interval           time.Duration
+	EpochEndFraction   float64
+	HareDelayLayers    uint32
+	SyncCertDistance   uint32
+	MaxStaleDuration   time.Duration
+	Standalone         bool
+	UseNewProtocol     bool `mapstructure:"use-new-opn"`
+	GossipDuration     time.Duration
+	OutOfSyncThreshold 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,
-		UseNewProtocol:   true,
-		GossipDuration:   15 * time.Second,
+		Interval:           10 * time.Second,
+		EpochEndFraction:   0.8,
+		HareDelayLayers:    10,
+		SyncCertDistance:   10,
+		MaxStaleDuration:   time.Second,
+		UseNewProtocol:     true,
+		GossipDuration:     15 * time.Second,
+		OutOfSyncThreshold: 3,
 	}
 }
 
-const (
-	outOfSyncThreshold uint32 = 3 // see notSynced
-)
-
 type syncState uint32
 
 const (
@@ -454,7 +452,7 @@ 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),
@@ -474,7 +472,7 @@ 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.OutOfSyncThreshold) {
 		s.setSyncState(ctx, notSynced)
 	}
 }
@@ -494,7 +492,7 @@ 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.OutOfSyncThreshold) {
 			s.setSyncState(ctx, notSynced)
 		}
 	case gossipSync:

diff --git a/syncer/syncer_test.go b/syncer/syncer_test.go
@@ -30,6 +30,8 @@ import (
 const (
 	layersPerEpoch = 3
 	never          = time.Second * 60 * 24
+
+	outOfSyncThreshold = 3
 )
 
 func TestMain(m *testing.M) {
@@ -95,12 +97,13 @@ 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,
-		UseNewProtocol:   true,
+		Interval:           interval,
+		GossipDuration:     5 * time.Millisecond,
+		EpochEndFraction:   0.66,
+		SyncCertDistance:   4,
+		HareDelayLayers:    5,
+		UseNewProtocol:     true,
+		OutOfSyncThreshold: outOfSyncThreshold,
 	}
 	ts.syncer = NewSyncer(ts.cdb, ts.mTicker, ts.mBeacon, ts.msh, nil, nil, ts.mLyrPatrol, ts.mCertHdr,
 		WithConfig(cfg),
@@ -539,7 +542,7 @@ 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