fix!: retrieve transactions from the mempool async, ignore irelevant rounds, and don't start processing messages until switching from blocksync to consensus

blockchain/v0/pool.go

@@ -32,7 +32,7 @@ const (
 	maxTotalRequesters        = 600
 	maxPendingRequests        = maxTotalRequesters
 	maxPendingRequestsPerPeer = 20
-	requestRetrySeconds       = 30
+	requestRetrySeconds       = 45
 
 	// Minimum recv rate to ensure we're receiving blocks from a peer fast
 	// enough. If a peer is not sending us data at least that rate, we
@@ -46,7 +46,7 @@ const (
 	maxDiffBetweenCurrentAndReceivedBlockHeight = 100
 )
 
-var peerTimeout = 15 * time.Second // not const so we can override with tests
+var peerTimeout = 120 * time.Second // not const so we can override with tests
 
 /*
 	Peers self report their heights when we join the block pool.

blockchain/v0/reactor.go

@@ -145,7 +145,7 @@ func (bcR *BlockchainReactor) GetChannels() []*p2p.ChannelDescriptor {
 	return []*p2p.ChannelDescriptor{
 		{
 			ID:                  BlockchainChannel,
-			Priority:            5,
+			Priority:            30,
 			SendQueueCapacity:   1000,
 			RecvBufferCapacity:  50 * 4096,
 			RecvMessageCapacity: bc.MaxMsgSize,

consensus/propagation/catchup.go

@@ -29,7 +29,7 @@ func (blockProp *Reactor) retryWants(currentHeight int64, currentRound int32) {
 		}
 
 		// only re-request original parts that are missing, not parity parts.
-		missing := prop.block.BitArray().Not()
+		missing := prop.block.MissingOriginal()
 		if missing.IsEmpty() {
 			blockProp.Logger.Error("no missing parts yet block is incomplete", "height", height, "round", round)
 			continue
@@ -40,6 +40,7 @@ func (blockProp *Reactor) retryWants(currentHeight int64, currentRound int32) {
 
 		for _, peer := range peers {
 			mc := missing.Copy()
+
 			reqs, has := peer.GetRequests(height, round)
 			if has {
 				mc = mc.Sub(reqs)
@@ -52,7 +53,7 @@ func (blockProp *Reactor) retryWants(currentHeight int64, currentRound int32) {
 			e := p2p.Envelope{
 				ChannelID: WantChannel,
 				Message: &protoprop.WantParts{
-					Parts:  *missing.ToProto(),
+					Parts:  *mc.ToProto(),
 					Height: height,
 					Round:  round,
 					Prove:  true,
@@ -72,9 +73,12 @@ func (blockProp *Reactor) retryWants(currentHeight int64, currentRound int32) {
 }
 
 func (blockProp *Reactor) AddCommitment(height int64, round int32, psh *types.PartSetHeader) {
+	blockProp.Logger.Info("adding commitment", "height", height, "round", round, "psh", psh)
 	blockProp.pmtx.Lock()
 	defer blockProp.pmtx.Unlock()
 
+	blockProp.Logger.Info("added commitment", "height", height, "round", round)
+
 	if blockProp.proposals[height] == nil {
 		blockProp.proposals[height] = make(map[int32]*proposalData)
 	}

consensus/propagation/catchup_test.go

@@ -1,75 +1,76 @@
 package propagation
 
-// TODO(rachid): fix test
-// func TestCatchup(t *testing.T) {
-// 	reactors, _ := testBlockPropReactors(3)
-// 	reactor1 := reactors[0]
-// 	reactor2 := reactors[1]
-// 	reactor3 := reactors[2]
-
-// 	// setting the proposal for height 8 round 1
-// 	compactBlock := createCompactBlock(8, 1)
-// 	reactor1.AddProposal(compactBlock)
-// 	reactor2.AddProposal(compactBlock)
-// 	reactor3.AddProposal(compactBlock)
-
-// 	// setting the proposal for height 9 round 0
-// 	compactBlock = createCompactBlock(9, 1)
-// 	reactor1.AddProposal(compactBlock)
-// 	reactor2.AddProposal(compactBlock)
-// 	reactor3.AddProposal(compactBlock)
-
-// 	// setting the proposal for height 10 round 0
-// 	compactBlock = createCompactBlock(10, 0)
-// 	reactor1.AddProposal(compactBlock)
-// 	reactor2.AddProposal(compactBlock)
-// 	reactor3.AddProposal(compactBlock)
-
-// 	// setting the proposal for height 10 round 1
-// 	compactBlock = createCompactBlock(10, 1)
-// 	reactor1.AddProposal(compactBlock)
-// 	reactor2.AddProposal(compactBlock)
-// 	reactor3.AddProposal(compactBlock)
-
-// 	// setting the first reactor current height and round
-// 	reactor1.currentHeight = 8
-// 	reactor1.currentRound = 0
-
-// 	// handle the compact block
-// 	reactor1.handleCompactBlock(compactBlock, reactor1.self)
-
-// 	time.Sleep(200 * time.Millisecond)
-
-// 	// check if reactor 1 sent wants to all the connected peers
-// 	_, has := reactor2.getPeer(reactor1.self).GetWants(9, 1)
-// 	require.True(t, has)
-
-// 	_, has = reactor2.getPeer(reactor1.self).GetWants(10, 0)
-// 	require.True(t, has)
-
-// 	_, has = reactor3.getPeer(reactor1.self).GetWants(9, 1)
-// 	require.True(t, has)
-
-// 	_, has = reactor3.getPeer(reactor1.self).GetWants(10, 0)
-// 	require.True(t, has)
-// }
-
-// func createCompactBlock(height int64, round int32) *proptypes.CompactBlock {
-// 	return &proptypes.CompactBlock{
-// 		BpHash:    cmtrand.Bytes(32),
-// 		Signature: cmtrand.Bytes(64),
-// 		LastLen:   0,
-// 		Blobs: []proptypes.TxMetaData{
-// 			{Hash: cmtrand.Bytes(32)},
-// 			{Hash: cmtrand.Bytes(32)},
-// 		},
-// 		Proposal: types.Proposal{
-// 			BlockID: types.BlockID{
-// 				Hash:          nil,
-// 				PartSetHeader: types.PartSetHeader{Total: 30},
-// 			},
-// 			Height: height,
-// 			Round:  round,
-// 		},
-// 	}
-// }
+import (
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+	cfg "github.com/tendermint/tendermint/config"
+	proptypes "github.com/tendermint/tendermint/consensus/propagation/types"
+	cmtrand "github.com/tendermint/tendermint/libs/rand"
+	"github.com/tendermint/tendermint/state"
+	"github.com/tendermint/tendermint/types"
+)
+
+func TestGapCatchup(t *testing.T) {
+	p2pCfg := cfg.DefaultP2PConfig()
+	p2pCfg.SendRate = 100000000
+	p2pCfg.RecvRate = 100000000
+	nodes := 2
+
+	reactors, _ := createTestReactors(nodes, p2pCfg, false, "/home/evan/data/experiments/celestia/fast-recovery/debug")
+	cleanup, _, sm := state.SetupTestCase(t)
+	t.Cleanup(func() {
+		cleanup(t)
+	})
+
+	prop, ps, _, metaData := createTestProposal(sm, 1, 2, 1000000)
+
+	// set the commitment and the data on the first node so that it can respond
+	// to the catchup request
+	n1 := reactors[0]
+	parityBlock, lastLen, err := types.Encode(ps, types.BlockPartSizeBytes)
+	require.NoError(t, err)
+
+	partHashes := extractHashes(ps, parityBlock)
+	proofs := extractProofs(ps, parityBlock)
+
+	cb := &proptypes.CompactBlock{
+		Proposal:    *prop,
+		LastLen:     uint32(lastLen),
+		Signature:   cmtrand.Bytes(64), // todo: sign the proposal with a real signature
+		BpHash:      parityBlock.Hash(),
+		Blobs:       metaData,
+		PartsHashes: partHashes,
+	}
+
+	cb.SetProofCache(proofs)
+
+	added := n1.AddProposal(cb)
+	require.True(t, added)
+
+	_, parts, _, has := n1.getAllState(prop.Height, prop.Round)
+	require.True(t, has)
+
+	parts.SetProposalData(ps, parityBlock)
+
+	// add the partset header to the second node and trigger the call to retry
+	// wants
+	n2 := reactors[1]
+
+	_, _, has = n2.GetProposal(prop.Height, prop.Round)
+	require.False(t, has)
+
+	psh := ps.Header()
+	n2.AddCommitment(prop.Height, prop.Round, &psh)
+
+	// this call simulates getting a commitment for a proposal of a higher
+	// height
+	n2.retryWants(2, 0)
+
+	time.Sleep(800 * time.Millisecond)
+
+	_, caughtUp, has := n2.GetProposal(prop.Height, prop.Round)
+	require.True(t, has)
+	require.True(t, caughtUp.IsComplete())
+}

consensus/propagation/commitment.go

@@ -1,7 +1,6 @@
 package propagation
 
 import (
-	"bytes"
 	"fmt"
 
 	"github.com/tendermint/tendermint/crypto/merkle"
@@ -45,7 +44,7 @@
 	cb.SetProofCache(proofs)
 
 	// save the compact block locally and broadcast it to the connected peers
-	blockProp.handleCompactBlock(&cb, blockProp.self)
+	blockProp.handleCompactBlock(&cb, blockProp.self, true)
 
 	_, parts, _, has := blockProp.getAllState(proposal.Height, proposal.Round)
 	if !has {
@@ -122,75 +121,36 @@
 // handleCompactBlock adds a proposal to the data routine. This should be called any
 // time a proposal is received from a peer or when a proposal is created. If the
 // proposal is new, it will be stored and broadcast to the relevant peers.
-func (blockProp *Reactor) handleCompactBlock(cb *proptypes.CompactBlock, peer p2p.ID) {
-	added, _, _ := blockProp.AddProposal(cb)
+func (blockProp *Reactor) handleCompactBlock(cb *proptypes.CompactBlock, peer p2p.ID, proposer bool) {
+	added := blockProp.AddProposal(cb)
 	if !added {
 		return
 	}
 
-	proofs, err := cb.Proofs()
+	// generate (and cache) the proofs from the partset hashes in the compact block
+	_, err := cb.Proofs()
 	if err != nil {
 		blockProp.DeleteRound(cb.Proposal.Height, cb.Proposal.Round)
 		blockProp.Logger.Error("received invalid compact block", "err", err.Error())
 		// todo: kick peer
 		return
 	}
 
-	// run the catchup routine to recover any missing parts for past heights.
-	blockProp.retryWants(cb.Proposal.Height, cb.Proposal.Round)
-
 	blockProp.broadcastCompactBlock(cb, peer)
 
-	// check if we have any transactions that are in the compact block
-	parts := blockProp.compactBlockToParts(cb)
-	_, partSet, found := blockProp.GetProposal(cb.Proposal.Height, cb.Proposal.Round)
-	if !found {
-		panic("failed to get proposal that was just added")
-	}
-
-	// the partset will be complete if this node is the proposer, and thus
-	// doesn't need to recover any parts.
-	if partSet.IsComplete() {
+	if proposer {
 		return
 	}
 
-	for _, part := range parts {
-		// todo: figure out what we want to do here. we might just want to defer
-		// to the consensus reactor for invalid parts.
-		if !bytes.Equal(merkle.LeafHash(part.Bytes), cb.PartsHashes[part.Index]) {
-			blockProp.Logger.Error(
-				"recovered part hash is different than compact block",
-				"part",
-				part.Index,
-				"height",
-				cb.Proposal.Height,
-				"round",
-				cb.Proposal.Round,
-			)
-			continue
-		}
-
-		part.Proof = *proofs[part.Index]
-
-		// note: using AddPartWithoutProof to skip verifying the proof that was
-		// generated and verified above.
-		added, err := partSet.AddPartWithoutProof(part)
-		if err != nil {
-			blockProp.Logger.Error("failed to add locally recovered part", "err", err)
-			continue
-		}
-
-		if !added {
-			blockProp.Logger.Error("failed to add locally recovered part", "part", part.Index)
-			continue
-		}
+	// run the catchup routine to recover any missing parts for past heights.
+	blockProp.retryWants(cb.Proposal.Height, cb.Proposal.Round)
 
-		// todo: broadcast haves
-	}
+	// check if we have any transactions that are in the compact block
+	go blockProp.recoverPartsFromMempool(cb)
 }
 
-// compactBlockToParts queries the mempool to see if we can recover any block parts locally.
-func (blockProp *Reactor) compactBlockToParts(cb *proptypes.CompactBlock) []*types.Part {
+// recoverPartsFromMempool queries the mempool to see if we can recover any block parts locally.
+func (blockProp *Reactor) recoverPartsFromMempool(cb *proptypes.CompactBlock) {
 	// find the compact block transactions that exist in our mempool
 	txsFound := make([]proptypes.UnmarshalledTx, 0)
 	for _, txMetaData := range cb.Blobs {
@@ -199,6 +159,7 @@
 			blockProp.Logger.Error("failed to decode tx key", "err", err, "tx", txMetaData)
 			continue
 		}
+
 		tx, has := blockProp.mempool.GetTxByKey(txKey)
 		if !has {
 			continue
@@ -213,17 +174,52 @@
 
 		txsFound = append(txsFound, proptypes.UnmarshalledTx{MetaData: txMetaData, Key: txKey, TxBytes: marshalledTx})
 	}
+
 	if len(txsFound) == 0 {
-		// no compact block transaction was found locally
-		return nil
+		return
 	}
 
 	parts := proptypes.TxsToParts(txsFound)
 	if len(parts) > 0 {
+		// todo: remove this log
 		blockProp.Logger.Info("recovered parts from the mempool", "number of parts", len(parts))
 	}
 
-	return parts
+	_, partSet, _, found := blockProp.getAllState(cb.Proposal.Height, cb.Proposal.Round)
+	if !found {
+		blockProp.Logger.Error("failed to get all state for this node's proposal", "height", cb.Proposal.Height, "round", cb.Proposal.Round)
+	}
+
+	proofs, err := cb.Proofs()
+	if err != nil {
+		blockProp.Logger.Error("failed to get proofs from compact block", "err", err)
+		return
+	}
+
+	originalParts := partSet.Original()
+	recoveredCount := 0
+	for _, p := range parts {
+		p.Proof = *proofs[p.Index]
+
+		// adding the part without checking the proof since the proof was just
+		// created above.
+		added, err := originalParts.AddPartWithoutProof(p)
+		if err != nil {
+			blockProp.Logger.Error("failed to add locally recovered part", "err", err)
+			continue
+		}
+
+		if !added {
+			blockProp.Logger.Error("failed to add locally recovered part", "part", p.Index)
+			continue
+		}
+
+		recoveredCount++
+	}
+
+	schema.WriteMempoolRecoveredParts(blockProp.traceClient, cb.Proposal.Height, cb.Proposal.Round, recoveredCount)
+
+	return
 }
 
 // broadcastProposal gossips the provided proposal to all peers. This should