Backport of #5377 for v1.2

config/mainnet.go

@@ -167,6 +167,7 @@ func MainnetConfig() Config {
 			Standalone:               false,
 			GossipDuration:           50 * time.Second,
 			OutOfSyncThresholdLayers: 36, // 3h
+			DisableAtxSync:           true,
 		},
 		Recovery: checkpoint.DefaultConfig(),
 		Cache:    datastore.DefaultConfig(),

node/node.go

@@ -70,6 +70,7 @@ import (
 	"github.com/spacemeshos/go-spacemesh/sql/layers"
 	dbmetrics "github.com/spacemeshos/go-spacemesh/sql/metrics"
 	"github.com/spacemeshos/go-spacemesh/syncer"
+	"github.com/spacemeshos/go-spacemesh/syncer/atxsync"
 	"github.com/spacemeshos/go-spacemesh/syncer/blockssync"
 	"github.com/spacemeshos/go-spacemesh/system"
 	"github.com/spacemeshos/go-spacemesh/timesync"
@@ -1102,6 +1103,17 @@ func (app *App) listenToUpdates(ctx context.Context) {
 				}
 				if len(update.Data.ActiveSet) > 0 {
 					app.hOracle.UpdateActiveSet(update.Data.Epoch, update.Data.ActiveSet)
+
+					app.eg.Go(func() error {
+						return atxsync.Download(
+							ctx,
+							10*time.Second,
+							app.addLogger(SyncLogger, app.log).Zap(),
+							app.db,
+							app.fetcher,
+							update.Data.ActiveSet,
+						)
+					})
 				}
 			}
 		}

syncer/atxsync/atxsync.go

@@ -0,0 +1,75 @@
+package atxsync
+
+import (
+	"context"
+	"math/rand"
+	"time"
+
+	"go.uber.org/zap"
+	"go.uber.org/zap/zapcore"
+
+	"github.com/spacemeshos/go-spacemesh/common/types"
+	"github.com/spacemeshos/go-spacemesh/sql"
+	"github.com/spacemeshos/go-spacemesh/sql/atxs"
+)
+
+//go:generate mockgen -typed -package=mocks -destination=./mocks/mocks.go -source=./atxsync.go
+type atxFetcher interface {
+	GetAtxs(context.Context, []types.ATXID) error
+}
+
+func getMissing(db *sql.Database, set []types.ATXID) ([]types.ATXID, error) {
+	missing := []types.ATXID{}
+	for _, atx := range set {
+		exist, err := atxs.Has(db, atx)
+		if err != nil {
+			return nil, err
+		}
+		if !exist {
+			missing = append(missing, atx)
+		}
+	}
+	return missing, nil
+}
+
+// Download specified set of atxs from peers in the network.
+//
+// actual retry interval will be between [retryInterval, 2*retryInterval]
+func Download(
+	ctx context.Context,
+	retryInterval time.Duration,
+	logger *zap.Logger,
+	db *sql.Database,
+	fetcher atxFetcher,
+	set []types.ATXID,
+) error {
+	total := len(set)
+	for {
+		missing, err := getMissing(db, set)
+		if err != nil {
+			return err
+		}
+		set = missing
+		downloaded := total - len(missing)
+		logger.Info("downloaded atxs",
+			zap.Int("total", total),
+			zap.Int("downloaded", downloaded),
+			zap.Array("missing", zapcore.ArrayMarshalerFunc(func(enc zapcore.ArrayEncoder) error {
+				for _, atx := range missing {
+					enc.AppendString(atx.ShortString())
+				}
+				return nil
+			})))
+		if downloaded == total {
+			return nil
+		}
+		if err := fetcher.GetAtxs(ctx, missing); err != nil {
+			logger.Debug("failed to fetch atxs", zap.Error(err))
+			select {
+			case <-ctx.Done():
+				return ctx.Err()
+			case <-time.After(retryInterval + time.Duration(rand.Int63n(int64(retryInterval)))):
+			}
+		}
+	}
+}

syncer/atxsync/atxsync_test.go

@@ -0,0 +1,122 @@
+package atxsync
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/spacemeshos/go-spacemesh/common/types"
+	"github.com/spacemeshos/go-spacemesh/log/logtest"
+	"github.com/spacemeshos/go-spacemesh/sql"
+	"github.com/spacemeshos/go-spacemesh/sql/atxs"
+	"github.com/spacemeshos/go-spacemesh/syncer/atxsync/mocks"
+	"github.com/stretchr/testify/require"
+	"go.uber.org/mock/gomock"
+)
+
+func atx(id types.ATXID) *types.VerifiedActivationTx {
+	atx := &types.ActivationTx{InnerActivationTx: types.InnerActivationTx{
+		NIPostChallenge: types.NIPostChallenge{
+			PublishEpoch: 1,
+		},
+		NumUnits: 1,
+	}}
+	atx.SetID(id)
+	atx.SetEffectiveNumUnits(1)
+	atx.SetReceived(time.Now())
+	copy(atx.SmesherID[:], id[:])
+	vatx, err := atx.Verify(0, 1)
+	if err != nil {
+		panic(err)
+	}
+	return vatx
+}
+
+func id(id ...byte) types.ATXID {
+	var atxid types.ATXID
+	copy(atxid[:], id)
+	return atxid
+}
+
+type fetchRequest struct {
+	request []types.ATXID
+	result  []*types.VerifiedActivationTx
+	error   error
+}
+
+func TestDownload(t *testing.T) {
+	canceled, cancel := context.WithCancel(context.Background())
+	cancel()
+	for _, tc := range []struct {
+		desc     string
+		ctx      context.Context
+		retry    time.Duration
+		existing []*types.VerifiedActivationTx
+		set      []types.ATXID
+		fetched  []fetchRequest
+		rst      error
+	}{
+		{
+			desc:     "all existing",
+			ctx:      context.Background(),
+			existing: []*types.VerifiedActivationTx{atx(id(1)), atx(id(2)), atx(id(3))},
+			set:      []types.ATXID{id(1), id(2), id(3)},
+		},
+		{
+			desc:     "with multiple requests",
+			ctx:      context.Background(),
+			existing: []*types.VerifiedActivationTx{atx(id(1))},
+			fetched: []fetchRequest{
+				{request: []types.ATXID{id(2), id(3)}, result: []*types.VerifiedActivationTx{atx(id(2))}},
+				{request: []types.ATXID{id(3)}, result: []*types.VerifiedActivationTx{atx(id(3))}},
+			},
+			set: []types.ATXID{id(1), id(2), id(3)},
+		},
+		{
+			desc:     "continue on error",
+			ctx:      context.Background(),
+			retry:    1,
+			existing: []*types.VerifiedActivationTx{atx(id(1))},
+			fetched: []fetchRequest{
+				{request: []types.ATXID{id(2)}, error: errors.New("test")},
+				{request: []types.ATXID{id(2)}, result: []*types.VerifiedActivationTx{atx(id(2))}},
+			},
+			set: []types.ATXID{id(1), id(2)},
+		},
+		{
+			desc:     "exit on context",
+			ctx:      canceled,
+			retry:    1,
+			existing: []*types.VerifiedActivationTx{atx(id(1))},
+			fetched: []fetchRequest{
+				{request: []types.ATXID{id(2)}, error: errors.New("test")},
+			},
+			set: []types.ATXID{id(1), id(2)},
+			rst: context.Canceled,
+		},
+	} {
+		t.Run(tc.desc, func(t *testing.T) {
+			logger := logtest.New(t)
+			db := sql.InMemory()
+			ctrl := gomock.NewController(t)
+			fetcher := mocks.NewMockatxFetcher(ctrl)
+			for _, atx := range tc.existing {
+				require.NoError(t, atxs.Add(db, atx))
+			}
+			for i := range tc.fetched {
+				req := tc.fetched[i]
+				fetcher.EXPECT().
+					GetAtxs(tc.ctx, req.request).
+					Times(1).
+					DoAndReturn(func(_ context.Context, _ []types.ATXID) error {
+						for _, atx := range req.result {
+							require.NoError(t, atxs.Add(db, atx))
+						}
+						return req.error
+					})
+			}
+			require.Equal(t, tc.rst, Download(tc.ctx, tc.retry, logger.Zap(), db, fetcher, tc.set))
+		})
+	}
+}

syncer/syncer.go

@@ -29,6 +29,7 @@ type Config struct {
 	Standalone               bool
 	GossipDuration           time.Duration
 	OutOfSyncThresholdLayers uint32 `mapstructure:"out-of-sync-threshold"`
+	DisableAtxSync           bool
 }
 
 // DefaultConfig for the syncer.
@@ -443,6 +444,11 @@ func (s *Syncer) syncAtx(ctx context.Context) error {
 		}
 	}
 
+	if s.cfg.DisableAtxSync {
+		s.logger.Debug("atx sync disabled")
+		return nil
+	}
+
 	// steady state atx syncing
 	curr := s.ticker.CurrentLayer()
 	if float64(