fix(miner): ignore lastWork when selecting the best mining candidate + tests

CHANGELOG.md

@@ -23,6 +23,10 @@
 - Change the F3 HeadLookback parameter to 4 ([filecoin-project/lotus#12648](https://github.com/filecoin-project/lotus/pull/12648)).
 - Upgrade go-f3 to 0.7.1 to resolve Tipset not found errors when trying to establish instance start time ([filecoin-project/lotus#12651](https://github.com/filecoin-project/lotus/pull/12651)).
 
+## Changes
+
+- The Lotus Miner will now always mine on the latest chain head returned by lotus, even if that head has less "weight" than the previously seen head. This is necessary because F3 may end up finalizing a tipset with a lower weight, although this situation should be rare on the Filecoin mainnet. ([filecoin-project/lotus#12659](https://github.com/filecoin-project/lotus/pull/12659))
+
 ## Deps
 
 # UNRELEASED Node v1.30.0  

chain/exchange/client.go

@@ -496,6 +496,9 @@ func (c *client) sendRequestToPeer(ctx context.Context, peer peer.ID, req *Reque
 
 // AddPeer implements Client.AddPeer(). Refer to the godocs there.
 func (c *client) AddPeer(p peer.ID) {
+	if p == c.host.ID() {
+		return
+	}
 	c.peerTracker.addPeer(p)
 }
 

itests/checkpoint_test.go

@@ -0,0 +1,64 @@
+package itests
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/filecoin-project/go-state-types/abi"
+	"github.com/filecoin-project/lotus/chain/types"
+	"github.com/filecoin-project/lotus/itests/kit"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCheckpointFork(t *testing.T) {
+	ctx := context.Background()
+
+	blocktime := 100 * time.Millisecond
+
+	nopts := []kit.NodeOpt{kit.WithAllSubsystems(), kit.ThroughRPC()}
+	var n1, n2 kit.TestFullNode
+	var m1, m2 kit.TestMiner
+	ens := kit.NewEnsemble(t, kit.MockProofs()).
+		FullNode(&n1, nopts...).
+		FullNode(&n2, nopts...).
+		Miner(&m1, &n1, nopts...).
+		Miner(&m2, &n1, nopts...).Start()
+
+	// Start 2 of them.
+	ens.InterconnectAll().BeginMining(blocktime)
+
+	{
+		ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+		n1.WaitTillChain(ctx, kit.HeightAtLeast(abi.ChainEpoch(5)))
+		cancel()
+	}
+
+	// Wait till both participate in a single tipset.
+	var target *types.TipSet
+	{
+		// find the first tipset where two miners mine a block.
+		ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
+		target = n1.WaitTillChain(ctx, func(ts *types.TipSet) bool {
+			return len(ts.Blocks()) == 2
+		})
+		cancel()
+	}
+
+	// Wait till we've moved on from that tipset.
+	targetHeight := target.Height() + 10
+	n1.WaitTillChain(ctx, kit.HeightAtLeast(targetHeight))
+	n2.WaitTillChain(ctx, kit.HeightAtLeast(targetHeight))
+
+	// Forcibly sync to this fork tipset.
+	forkTs, err := types.NewTipSet(target.Blocks()[:1])
+	require.NoError(t, err)
+	require.NoError(t, n2.SyncCheckpoint(ctx, forkTs.Key()))
+	require.NoError(t, n1.SyncCheckpoint(ctx, forkTs.Key()))
+
+	// See if we can start making progress again!
+	newHead := n1.WaitTillChain(ctx, kit.HeightAtLeast(targetHeight))
+	forkTs2, err := n1.ChainGetTipSetByHeight(ctx, forkTs.Height(), newHead.Key())
+	require.NoError(t, err)
+	require.True(t, forkTs.Equals(forkTs2))
+}

miner/miner.go

@@ -311,7 +311,8 @@ minerLoop:
 		onDone(b != nil, h, nil)
 
 		// Process the mined block.
-		if b != nil {
+		switch {
+		case b != nil:
 			// Record the event of mining a block.
 			m.journal.RecordEvent(m.evtTypes[evtTypeBlockMined], func() interface{} {
 				return map[string]interface{}{
@@ -344,23 +345,23 @@ minerLoop:
 
 			// Check for slash filter conditions.
 			if os.Getenv("LOTUS_MINER_NO_SLASHFILTER") != "_yes_i_know_i_can_and_probably_will_lose_all_my_fil_and_power_" && !buildconstants.IsNearUpgrade(base.TipSet.Height(), buildconstants.UpgradeWatermelonFixHeight) {
-				witness, fault, err := m.sf.MinedBlock(ctx, b.Header, base.TipSet.Height()+base.NullRounds)
+				witness, fault, err := m.sf.MinedBlock(ctx, b.Header, base.TipSet.Height())
 				if err != nil {
 					log.Errorf("<!!> SLASH FILTER ERRORED: %s", err)
 					// Continue here, because it's _probably_ wiser to not submit this block
-					continue
+					break
 				}
 
 				if fault {
 					log.Errorf("<!!> SLASH FILTER DETECTED FAULT due to blocks %s and %s", b.Header.Cid(), witness)
-					continue
+					break
 				}
 			}
 
 			// Check for blocks created at the same height.
 			if _, ok := m.minedBlockHeights.Get(b.Header.Height); ok {
 				log.Warnw("Created a block at the same height as another block we've created", "height", b.Header.Height, "miner", b.Header.Miner, "parents", b.Header.Parents)
-				continue
+				break
 			}
 
 			// Add the block height to the mined block heights.
@@ -369,24 +370,26 @@ minerLoop:
 			// Submit the newly mined block.
 			if err := m.api.SyncSubmitBlock(ctx, b); err != nil {
 				log.Errorf("failed to submit newly mined block: %+v", err)
+				break
 			}
-		} else {
-			// If no block was mined, increase the null rounds and wait for the next epoch.
-			base.NullRounds++
-
-			// Calculate the time for the next round.
-			nextRound := time.Unix(int64(base.TipSet.MinTimestamp()+buildconstants.BlockDelaySecs*uint64(base.NullRounds))+int64(buildconstants.PropagationDelaySecs), 0)
-
-			// Wait for the next round or stop signal.
-			select {
-			case <-build.Clock.After(build.Clock.Until(nextRound)):
-			case <-m.stop:
-				stopping := m.stopping
-				m.stop = nil
-				m.stopping = nil
-				close(stopping)
-				return
-			}
+			continue // TODO: we should probably remove this continue and wait in this case as well... but that's a bigger change.
+		}
+
+		// If no block was mined or if we fail to submit the block, increase the null rounds and wait for the next epoch.
+		base.NullRounds++
+
+		// Calculate the time for the next round.
+		nextRound := time.Unix(int64(base.TipSet.MinTimestamp()+buildconstants.BlockDelaySecs*uint64(base.NullRounds))+int64(buildconstants.PropagationDelaySecs), 0)
+
+		// Wait for the next round or stop signal.
+		select {
+		case <-build.Clock.After(build.Clock.Until(nextRound)):
+		case <-m.stop:
+			stopping := m.stopping
+			m.stop = nil
+			m.stopping = nil
+			close(stopping)
+			return
 		}
 	}
 }
@@ -400,11 +403,8 @@ type MiningBase struct {
 }
 
 // GetBestMiningCandidate implements the fork choice rule from a miner's
-// perspective.
-//
-// It obtains the current chain head (HEAD), and compares it to the last tipset
-// we selected as our mining base (LAST). If HEAD's weight is larger than
-// LAST's weight, it selects HEAD to build on. Else, it selects LAST.
+// perspective, returning the best head to mine on. This includes the number of null rounds we think
+// we should insert and the time at which we received said head.
 func (m *Miner) GetBestMiningCandidate(ctx context.Context) (*MiningBase, error) {
 	m.lk.Lock()
 	defer m.lk.Unlock()
@@ -414,27 +414,10 @@ func (m *Miner) GetBestMiningCandidate(ctx context.Context) (*MiningBase, error)
 		return nil, err
 	}
 
-	if m.lastWork != nil {
-		if m.lastWork.TipSet.Equals(bts) {
-			return m.lastWork, nil
-		}
-
-		btsw, err := m.api.ChainTipSetWeight(ctx, bts.Key())
-		if err != nil {
-			return nil, err
-		}
-		ltsw, err := m.api.ChainTipSetWeight(ctx, m.lastWork.TipSet.Key())
-		if err != nil {
-			m.lastWork = nil
-			return nil, err
-		}
-
-		if types.BigCmp(btsw, ltsw) <= 0 {
-			return m.lastWork, nil
-		}
+	if m.lastWork == nil || !m.lastWork.TipSet.Equals(bts) {
+		m.lastWork = &MiningBase{TipSet: bts, ComputeTime: time.Now()}
 	}
 
-	m.lastWork = &MiningBase{TipSet: bts, ComputeTime: time.Now()}
 	return m.lastWork, nil
 }
 

node/hello/hello.go

@@ -132,13 +132,11 @@ func (hs *Service) HandleStream(s inet.Stream) {
 		return
 	}
 
-	if ts.TipSet().Height() > 0 {
-		hs.h.ConnManager().TagPeer(s.Conn().RemotePeer(), "fcpeer", 10)
+	hs.h.ConnManager().TagPeer(s.Conn().RemotePeer(), "fcpeer", 10)
 
-		// don't bother informing about genesis
-		log.Debugf("Got new tipset through Hello: %s from %s", ts.Cids(), s.Conn().RemotePeer())
-		hs.syncer.InformNewHead(s.Conn().RemotePeer(), ts)
-	}
+	// don't bother informing about genesis
+	log.Debugf("Got new tipset through Hello: %s from %s", ts.Cids(), s.Conn().RemotePeer())
+	hs.syncer.InformNewHead(s.Conn().RemotePeer(), ts)
 }
 
 func (hs *Service) SayHello(ctx context.Context, pid peer.ID) error {