Merge pull request #3602 from filecoin-project/feat/delayed-pubsub-join

only subscribe to pubsub topics once we are synced

chain/store/store.go

@@ -5,6 +5,7 @@ import (
 	"context"
 	"encoding/binary"
 	"encoding/json"
+	"errors"
 	"io"
 	"os"
 	"strconv"
@@ -51,6 +52,8 @@ var blockValidationCacheKeyPrefix = dstore.NewKey("blockValidation")
 var DefaultTipSetCacheSize = 8192
 var DefaultMsgMetaCacheSize = 2048
 
+var ErrNotifieeDone = errors.New("notifee is done and should be removed")
+
 func init() {
 	if s := os.Getenv("LOTUS_CHAIN_TIPSET_CACHE"); s != "" {
 		tscs, err := strconv.Atoi(s)
@@ -358,11 +361,33 @@ func (cs *ChainStore) reorgWorker(ctx context.Context, initialNotifees []ReorgNo
 					apply[i], apply[opp] = apply[opp], apply[i]
 				}
 
-				for _, hcf := range notifees {
+				var toremove map[int]struct{}
-					if err := hcf(revert, apply); err != nil {
+				for i, hcf := range notifees {
-						log.Error("head change func errored (BAD): ", err)
+					err := hcf(revert, apply)
+					if err != nil {
+						if err == ErrNotifieeDone {
+							if toremove == nil {
+								toremove = make(map[int]struct{})
+							}
+							toremove[i] = struct{}{}
+						} else {
+							log.Error("head change func errored (BAD): ", err)
+						}
 					}
 				}
+
+				if len(toremove) > 0 {
+					newNotifees := make([]ReorgNotifee, 0, len(notifees)-len(toremove))
+					for i, hcf := range notifees {
+						_, remove := toremove[i]
+						if remove {
+							continue
+						}
+						newNotifees = append(newNotifees, hcf)
+					}
+					notifees = newNotifees
+				}
+
 			case <-ctx.Done():
 				return
 			}

chain/sync_test.go

@@ -32,6 +32,7 @@ import (
 	"github.com/filecoin-project/lotus/node"
 	"github.com/filecoin-project/lotus/node/impl"
 	"github.com/filecoin-project/lotus/node/modules"
+	"github.com/filecoin-project/lotus/node/modules/dtypes"
 	"github.com/filecoin-project/lotus/node/repo"
 )
 
@@ -231,6 +232,7 @@ func (tu *syncTestUtil) addSourceNode(gen int) {
 		node.Repo(sourceRepo),
 		node.MockHost(tu.mn),
 		node.Test(),
+		node.Override(new(dtypes.Bootstrapper), dtypes.Bootstrapper(true)),
 
 		node.Override(new(modules.Genesis), modules.LoadGenesis(genesis)),
 	)
@@ -263,6 +265,7 @@ func (tu *syncTestUtil) addClientNode() int {
 		node.Repo(repo.NewMemory(nil)),
 		node.MockHost(tu.mn),
 		node.Test(),
+		node.Override(new(dtypes.Bootstrapper), dtypes.Bootstrapper(true)),
 
 		node.Override(new(modules.Genesis), modules.LoadGenesis(tu.genesis)),
 	)

node/modules/services.go

@@ -1,6 +1,8 @@
 package modules
 
 import (
+	"time"
+
 	"github.com/ipfs/go-datastore"
 	"github.com/ipfs/go-datastore/namespace"
 	eventbus "github.com/libp2p/go-eventbus"
@@ -22,6 +24,7 @@ import (
 	"github.com/filecoin-project/lotus/chain/stmgr"
 	"github.com/filecoin-project/lotus/chain/store"
 	"github.com/filecoin-project/lotus/chain/sub"
+	"github.com/filecoin-project/lotus/chain/types"
 	"github.com/filecoin-project/lotus/lib/peermgr"
 	"github.com/filecoin-project/lotus/node/hello"
 	"github.com/filecoin-project/lotus/node/modules/dtypes"
@@ -73,14 +76,45 @@ func RunBlockSync(h host.Host, svc *blocksync.BlockSyncService) {
 	h.SetStreamHandler(blocksync.BlockSyncProtocolID, svc.HandleStream)
 }
 
-func HandleIncomingBlocks(mctx helpers.MetricsCtx, lc fx.Lifecycle, ps *pubsub.PubSub, s *chain.Syncer, bserv dtypes.ChainBlockService, chain *store.ChainStore, stmgr *stmgr.StateManager, h host.Host, nn dtypes.NetworkName) {
+func waitForSync(stmgr *stmgr.StateManager, epochs int, subscribe func()) {
-	ctx := helpers.LifecycleCtx(mctx, lc)
+	nearsync := uint64(epochs) * uint64(build.BlockDelaySecs) * uint64(time.Second) //nolint
 
-	blocksub, err := ps.Subscribe(build.BlocksTopic(nn)) //nolint
+	// early check, are we synced at start up?
-	if err != nil {
+	ts := stmgr.ChainStore().GetHeaviestTipSet()
-		panic(err)
+	timestamp := ts.MinTimestamp()
+	now := uint64(build.Clock.Now().UnixNano())
+	if timestamp > now-nearsync {
+		subscribe()
+		return
 	}
 
+	// we are not synced, subscribe to head changes and wait for sync
+	stmgr.ChainStore().SubscribeHeadChanges(func(rev, app []*types.TipSet) error {
+		if len(app) == 0 {
+			return nil
+		}
+
+		latest := app[0].MinTimestamp()
+		for _, ts := range app[1:] {
+			timestamp := ts.MinTimestamp()
+			if timestamp > latest {
+				latest = timestamp
+			}
+		}
+
+		now := uint64(build.Clock.Now().UnixNano())
+		if latest > now-nearsync {
+			subscribe()
+			return store.ErrNotifieeDone
+		}
+
+		return nil
+	})
+}
+
+func HandleIncomingBlocks(mctx helpers.MetricsCtx, lc fx.Lifecycle, ps *pubsub.PubSub, s *chain.Syncer, bserv dtypes.ChainBlockService, chain *store.ChainStore, stmgr *stmgr.StateManager, h host.Host, nn dtypes.NetworkName, bootstrapper dtypes.Bootstrapper) {
+	ctx := helpers.LifecycleCtx(mctx, lc)
+
 	v := sub.NewBlockValidator(
 		h.ID(), chain, stmgr,
 		func(p peer.ID) {
@@ -92,24 +126,53 @@ func HandleIncomingBlocks(mctx helpers.MetricsCtx, lc fx.Lifecycle, ps *pubsub.P
 		panic(err)
 	}
 
-	go sub.HandleIncomingBlocks(ctx, blocksub, s, bserv, h.ConnManager())
+	subscribe := func() {
+		log.Infof("subscribing to pubsub topic %s", build.BlocksTopic(nn))
+
+		blocksub, err := ps.Subscribe(build.BlocksTopic(nn)) //nolint
+		if err != nil {
+			panic(err)
+		}
+
+		go sub.HandleIncomingBlocks(ctx, blocksub, s, bserv, h.ConnManager())
+	}
+
+	if bootstrapper {
+		subscribe()
+		return
+	}
+
+	// wait until we are synced within 10 blocks
+	waitForSync(stmgr, 10, subscribe)
 }
 
-func HandleIncomingMessages(mctx helpers.MetricsCtx, lc fx.Lifecycle, ps *pubsub.PubSub, mpool *messagepool.MessagePool, h host.Host, nn dtypes.NetworkName) {
+func HandleIncomingMessages(mctx helpers.MetricsCtx, lc fx.Lifecycle, ps *pubsub.PubSub, stmgr *stmgr.StateManager, mpool *messagepool.MessagePool, h host.Host, nn dtypes.NetworkName, bootstrapper dtypes.Bootstrapper) {
 	ctx := helpers.LifecycleCtx(mctx, lc)
 
-	msgsub, err := ps.Subscribe(build.MessagesTopic(nn)) //nolint:staticcheck
-	if err != nil {
-		panic(err)
-	}
-
 	v := sub.NewMessageValidator(h.ID(), mpool)
 
 	if err := ps.RegisterTopicValidator(build.MessagesTopic(nn), v.Validate); err != nil {
 		panic(err)
 	}
 
-	go sub.HandleIncomingMessages(ctx, mpool, msgsub)
+	subscribe := func() {
+		log.Infof("subscribing to pubsub topic %s", build.MessagesTopic(nn))
+
+		msgsub, err := ps.Subscribe(build.MessagesTopic(nn)) //nolint
+		if err != nil {
+			panic(err)
+		}
+
+		go sub.HandleIncomingMessages(ctx, mpool, msgsub)
+	}
+
+	if bootstrapper {
+		subscribe()
+		return
+	}
+
+	// wait until we are synced within 1 block
+	waitForSync(stmgr, 1, subscribe)
 }
 
 func NewLocalDiscovery(ds dtypes.MetadataDS) *discovery.Local {

node/test/builder.go

@@ -31,6 +31,7 @@ import (
 	miner2 "github.com/filecoin-project/lotus/miner"
 	"github.com/filecoin-project/lotus/node"
 	"github.com/filecoin-project/lotus/node/modules"
+	"github.com/filecoin-project/lotus/node/modules/dtypes"
 	testing2 "github.com/filecoin-project/lotus/node/modules/testing"
 	"github.com/filecoin-project/lotus/node/repo"
 	"github.com/filecoin-project/lotus/storage/mockstorage"
@@ -370,6 +371,8 @@ func MockSbBuilder(t *testing.T, nFull int, storage []test.StorageMiner) ([]test
 
 			node.Override(new(ffiwrapper.Verifier), mock.MockVerifier),
 
+			node.Override(new(dtypes.Bootstrapper), dtypes.Bootstrapper(true)),
+
 			genesis,
 		)
 		if err != nil {