Merge branch 'utxo-child-bug' into bump_v0.12.4-rc1

* Logger: change log level for "Couldn't find UTXO entry" to debug

* ignore error for orphans

* continue also if orphans

* check if blocks exist

* safe rpc mode

* limit window size

* verify block status depending on context

* allow a 2 factor gap in expected mergeset size

Co-authored-by: msutton <mikisiton2@gmail.com>

app/protocol/flows/v5/blockrelay/handle_ibd_block_locator.go

@@ -5,7 +5,6 @@ import (
 	"github.com/kaspanet/kaspad/app/protocol/peer"
 	"github.com/kaspanet/kaspad/app/protocol/protocolerrors"
 	"github.com/kaspanet/kaspad/domain"
-	"github.com/kaspanet/kaspad/domain/consensus/model/externalapi"
 	"github.com/kaspanet/kaspad/infrastructure/network/netadapter/router"
 )
 
@@ -34,7 +33,7 @@ func HandleIBDBlockLocator(context HandleIBDBlockLocatorContext, incomingRoute *
 		if err != nil {
 			return err
 		}
-		if !blockInfo.Exists {
+		if !blockInfo.HasHeader() {
 			return protocolerrors.Errorf(true, "received IBDBlockLocator "+
 				"with an unknown targetHash %s", targetHash)
 		}
@@ -47,7 +46,7 @@ func HandleIBDBlockLocator(context HandleIBDBlockLocatorContext, incomingRoute *
 			}
 
 			// The IBD block locator is checking only existing blocks with bodies.
-			if !blockInfo.Exists || blockInfo.BlockStatus == externalapi.StatusHeaderOnly {
+			if !blockInfo.HasBody() {
 				continue
 			}
 

app/protocol/flows/v5/blockrelay/handle_ibd_block_requests.go

@@ -4,7 +4,6 @@ import (
 	"github.com/kaspanet/kaspad/app/appmessage"
 	"github.com/kaspanet/kaspad/app/protocol/protocolerrors"
 	"github.com/kaspanet/kaspad/domain"
-	"github.com/kaspanet/kaspad/domain/consensus/model/externalapi"
 	"github.com/kaspanet/kaspad/infrastructure/network/netadapter/router"
 	"github.com/pkg/errors"
 )
@@ -32,7 +31,7 @@ func HandleIBDBlockRequests(context HandleIBDBlockRequestsContext, incomingRoute
 			if err != nil {
 				return err
 			}
-			if !blockInfo.Exists || blockInfo.BlockStatus == externalapi.StatusHeaderOnly {
+			if !blockInfo.HasBody() {
 				return protocolerrors.Errorf(true, "block %s not found (v5)", hash)
 			}
 			block, err := context.Domain().Consensus().GetBlock(hash)

app/protocol/flows/v5/blockrelay/handle_relay_block_requests.go

@@ -5,7 +5,6 @@ import (
 	peerpkg "github.com/kaspanet/kaspad/app/protocol/peer"
 	"github.com/kaspanet/kaspad/app/protocol/protocolerrors"
 	"github.com/kaspanet/kaspad/domain"
-	"github.com/kaspanet/kaspad/domain/consensus/model/externalapi"
 	"github.com/kaspanet/kaspad/infrastructure/network/netadapter/router"
 	"github.com/pkg/errors"
 )
@@ -33,7 +32,7 @@ func HandleRelayBlockRequests(context RelayBlockRequestsContext, incomingRoute *
 			if err != nil {
 				return err
 			}
-			if !blockInfo.Exists || blockInfo.BlockStatus == externalapi.StatusHeaderOnly {
+			if !blockInfo.HasBody() {
 				return protocolerrors.Errorf(true, "block %s not found", hash)
 			}
 			block, err := context.Domain().Consensus().GetBlock(hash)

app/protocol/flows/v5/blockrelay/handle_request_anticone.go

@@ -47,9 +47,9 @@ func (flow *handleRequestAnticoneFlow) start() error {
 
 		// GetAnticone is expected to be called by the syncee for getting the anticone of the header selected tip
 		// intersected by past of relayed block, and is thus expected to be bounded by mergeset limit since
-		// we relay blocks only if they enter virtual's mergeset. We add 2 for a small margin error.
+		// we relay blocks only if they enter virtual's mergeset. We add a 2 factor for possible sync gaps.
 		blockHashes, err := flow.Domain().Consensus().GetAnticone(blockHash, contextHash,
-			flow.Config().ActiveNetParams.MergeSetSizeLimit+2)
+			flow.Config().ActiveNetParams.MergeSetSizeLimit*2)
 		if err != nil {
 			return protocolerrors.Wrap(true, err, "Failed querying anticone")
 		}

app/protocol/flows/v5/blockrelay/handle_request_headers.go

@@ -46,7 +46,25 @@ func (flow *handleRequestHeadersFlow) start() error {
 		}
 		log.Debugf("Received requestHeaders with lowHash: %s, highHash: %s", lowHash, highHash)
 
-		isLowSelectedAncestorOfHigh, err := flow.Domain().Consensus().IsInSelectedParentChainOf(lowHash, highHash)
+		consensus := flow.Domain().Consensus()
+
+		lowHashInfo, err := consensus.GetBlockInfo(lowHash)
+		if err != nil {
+			return err
+		}
+		if !lowHashInfo.HasHeader() {
+			return protocolerrors.Errorf(true, "Block %s does not exist", lowHash)
+		}
+
+		highHashInfo, err := consensus.GetBlockInfo(highHash)
+		if err != nil {
+			return err
+		}
+		if !highHashInfo.HasHeader() {
+			return protocolerrors.Errorf(true, "Block %s does not exist", highHash)
+		}
+
+		isLowSelectedAncestorOfHigh, err := consensus.IsInSelectedParentChainOf(lowHash, highHash)
 		if err != nil {
 			return err
 		}
@@ -62,7 +80,7 @@ func (flow *handleRequestHeadersFlow) start() error {
 			// in order to avoid locking the consensus for too long
 			// maxBlocks MUST be >= MergeSetSizeLimit + 1
 			const maxBlocks = 1 << 10
-			blockHashes, _, err := flow.Domain().Consensus().GetHashesBetween(lowHash, highHash, maxBlocks)
+			blockHashes, _, err := consensus.GetHashesBetween(lowHash, highHash, maxBlocks)
 			if err != nil {
 				return err
 			}
@@ -70,7 +88,7 @@ func (flow *handleRequestHeadersFlow) start() error {
 
 			blockHeaders := make([]*appmessage.MsgBlockHeader, len(blockHashes))
 			for i, blockHash := range blockHashes {
-				blockHeader, err := flow.Domain().Consensus().GetBlockHeader(blockHash)
+				blockHeader, err := consensus.GetBlockHeader(blockHash)
 				if err != nil {
 					return err
 				}

app/protocol/flows/v5/blockrelay/ibd.go

@@ -686,37 +686,28 @@ func (flow *handleIBDFlow) banIfBlockIsHeaderOnly(block *externalapi.DomainBlock
 }
 
 func (flow *handleIBDFlow) resolveVirtual(estimatedVirtualDAAScoreTarget uint64) error {
-	virtualDAAScoreStart, err := flow.Domain().Consensus().GetVirtualDAAScore()
+	err := flow.Domain().Consensus().ResolveVirtual(func(virtualDAAScoreStart uint64, virtualDAAScore uint64) {
+		var percents int
+		if estimatedVirtualDAAScoreTarget-virtualDAAScoreStart <= 0 {
+			percents = 100
+		} else {
+			percents = int(float64(virtualDAAScore-virtualDAAScoreStart) / float64(estimatedVirtualDAAScoreTarget-virtualDAAScoreStart) * 100)
+		}
+		if percents < 0 {
+			percents = 0
+		} else if percents > 100 {
+			percents = 100
+		}
+		log.Infof("Resolving virtual. Estimated progress: %d%%", percents)
+	})
 	if err != nil {
 		return err
 	}
 
-	for i := 0; ; i++ {
-		if i%10 == 0 {
-			virtualDAAScore, err := flow.Domain().Consensus().GetVirtualDAAScore()
-			if err != nil {
-				return err
-			}
-			var percents int
-			if estimatedVirtualDAAScoreTarget-virtualDAAScoreStart <= 0 {
-				percents = 100
-			} else {
-				percents = int(float64(virtualDAAScore-virtualDAAScoreStart) / float64(estimatedVirtualDAAScoreTarget-virtualDAAScoreStart) * 100)
-			}
-			log.Infof("Resolving virtual. Estimated progress: %d%%", percents)
-		}
-		isCompletelyResolved, err := flow.Domain().Consensus().ResolveVirtual()
-		if err != nil {
-			return err
-		}
-
-		if isCompletelyResolved {
-			log.Infof("Resolved virtual")
-			err = flow.OnNewBlockTemplate()
-			if err != nil {
-				return err
-			}
-			return nil
-		}
+	log.Infof("Resolved virtual")
+	err = flow.OnNewBlockTemplate()
+	if err != nil {
+		return err
 	}
+	return nil
 }

app/rpc/rpchandlers/add_peer.go

@@ -9,6 +9,14 @@ import (
 
 // HandleAddPeer handles the respectively named RPC command
 func HandleAddPeer(context *rpccontext.Context, _ *router.Router, request appmessage.Message) (appmessage.Message, error) {
+	if context.Config.SafeRPC {
+		log.Warn("AddPeer RPC command called while node in safe RPC mode -- ignoring.")
+		response := appmessage.NewAddPeerResponseMessage()
+		response.Error =
+			appmessage.RPCErrorf("AddPeer RPC command called while node in safe RPC mode")
+		return response, nil
+	}
+
 	AddPeerRequest := request.(*appmessage.AddPeerRequestMessage)
 	address, err := network.NormalizeAddress(AddPeerRequest.Address, context.Config.ActiveNetParams.DefaultPort)
 	if err != nil {

app/rpc/rpchandlers/ban.go

@@ -9,6 +9,14 @@ import (
 
 // HandleBan handles the respectively named RPC command
 func HandleBan(context *rpccontext.Context, _ *router.Router, request appmessage.Message) (appmessage.Message, error) {
+	if context.Config.SafeRPC {
+		log.Warn("Ban RPC command called while node in safe RPC mode -- ignoring.")
+		response := appmessage.NewBanResponseMessage()
+		response.Error =
+			appmessage.RPCErrorf("Ban RPC command called while node in safe RPC mode")
+		return response, nil
+	}
+
 	banRequest := request.(*appmessage.BanRequestMessage)
 	ip := net.ParseIP(banRequest.IP)
 	if ip == nil {

app/rpc/rpchandlers/estimate_network_hashes_per_second.go

@@ -27,6 +27,27 @@ func HandleEstimateNetworkHashesPerSecond(
 		}
 	}
 
+	if context.Config.SafeRPC {
+		const windowSizeLimit = 10000
+		if windowSize > windowSizeLimit {
+			response := &appmessage.EstimateNetworkHashesPerSecondResponseMessage{}
+			response.Error =
+				appmessage.RPCErrorf(
+					"Requested window size %d is larger than max allowed in RPC safe mode (%d)",
+					windowSize, windowSizeLimit)
+			return response, nil
+		}
+	}
+
+	if uint64(windowSize) > context.Config.ActiveNetParams.PruningDepth() {
+		response := &appmessage.EstimateNetworkHashesPerSecondResponseMessage{}
+		response.Error =
+			appmessage.RPCErrorf(
+				"Requested window size %d is larger than pruning point depth %d",
+				windowSize, context.Config.ActiveNetParams.PruningDepth())
+		return response, nil
+	}
+
 	networkHashesPerSecond, err := context.Domain.Consensus().EstimateNetworkHashesPerSecond(startHash, windowSize)
 	if err != nil {
 		response := &appmessage.EstimateNetworkHashesPerSecondResponseMessage{}

app/rpc/rpchandlers/get_mempool_entries_by_addresses.go

@@ -43,7 +43,7 @@ func HandleGetMempoolEntriesByAddresses(context *rpccontext.Context, _ *router.R
 	if getMempoolEntriesByAddressesRequest.IncludeOrphanPool {
 
 		orphanPoolTransactions := context.Domain.MiningManager().AllOrphanTransactions()
-		orphanPoolEntriesByAddresse, err := extractMempoolEntriesByAddressesFromTransactions(
+		orphanPoolEntriesByAddress, err := extractMempoolEntriesByAddressesFromTransactions(
 			context,
 			getMempoolEntriesByAddressesRequest.Addresses,
 			orphanPoolTransactions,
@@ -59,7 +59,7 @@ func HandleGetMempoolEntriesByAddresses(context *rpccontext.Context, _ *router.R
 			return errorMessage, nil
 		}
 
-		mempoolEntriesByAddresses = append(mempoolEntriesByAddresses, orphanPoolEntriesByAddresse...)
+		mempoolEntriesByAddresses = append(mempoolEntriesByAddresses, orphanPoolEntriesByAddress...)
 	}
 
 	return appmessage.NewGetMempoolEntriesByAddressesResponseMessage(mempoolEntriesByAddresses), nil
@@ -80,9 +80,13 @@ func extractMempoolEntriesByAddressesFromTransactions(context *rpccontext.Contex
 		for _, transaction := range transactions {
 
 			for i, input := range transaction.Inputs {
-				// TODO: Fix this
 				if input.UTXOEntry == nil {
-					log.Errorf("Couldn't find UTXO entry for input %d in mempool transaction %s. This is a bug and should be fixed.", i, consensushashing.TransactionID(transaction))
+					if !areOrphans { // Orphans can legitimately have `input.UTXOEntry == nil`
+						// TODO: Fix the underlying cause of the bug for non-orphan entries
+						log.Debugf(
+							"Couldn't find UTXO entry for input %d in mempool transaction %s. This is a bug and should be fixed.",
+							i, consensushashing.TransactionID(transaction))
+					}
 					continue
 				}
 

app/rpc/rpchandlers/resolve_finality_conflict.go

@@ -8,6 +8,14 @@ import (
 
 // HandleResolveFinalityConflict handles the respectively named RPC command
 func HandleResolveFinalityConflict(context *rpccontext.Context, _ *router.Router, request appmessage.Message) (appmessage.Message, error) {
+	if context.Config.SafeRPC {
+		log.Warn("ResolveFinalityConflict RPC command called while node in safe RPC mode -- ignoring.")
+		response := &appmessage.ResolveFinalityConflictResponseMessage{}
+		response.Error =
+			appmessage.RPCErrorf("ResolveFinalityConflict RPC command called while node in safe RPC mode")
+		return response, nil
+	}
+
 	response := &appmessage.ResolveFinalityConflictResponseMessage{}
 	response.Error = appmessage.RPCErrorf("not implemented")
 	return response, nil

app/rpc/rpchandlers/shut_down.go

@@ -12,6 +12,14 @@ const pauseBeforeShutDown = time.Second
 
 // HandleShutDown handles the respectively named RPC command
 func HandleShutDown(context *rpccontext.Context, _ *router.Router, _ appmessage.Message) (appmessage.Message, error) {
+	if context.Config.SafeRPC {
+		log.Warn("ShutDown RPC command called while node in safe RPC mode -- ignoring.")
+		response := appmessage.NewShutDownResponseMessage()
+		response.Error =
+			appmessage.RPCErrorf("ShutDown RPC command called while node in safe RPC mode")
+		return response, nil
+	}
+
 	log.Warn("ShutDown RPC called.")
 
 	// Wait a second before shutting down, to allow time to return the response to the caller

app/rpc/rpchandlers/unban.go

@@ -9,6 +9,14 @@ import (
 
 // HandleUnban handles the respectively named RPC command
 func HandleUnban(context *rpccontext.Context, _ *router.Router, request appmessage.Message) (appmessage.Message, error) {
+	if context.Config.SafeRPC {
+		log.Warn("Unban RPC command called while node in safe RPC mode -- ignoring.")
+		response := appmessage.NewUnbanResponseMessage()
+		response.Error =
+			appmessage.RPCErrorf("Unban RPC command called while node in safe RPC mode")
+		return response, nil
+	}
+
 	unbanRequest := request.(*appmessage.UnbanRequestMessage)
 	ip := net.ParseIP(unbanRequest.IP)
 	if ip == nil {

changelog.txt

@@ -1,3 +1,8 @@
+Kaspad v0.12.3 - 2022-06-29
+===========================
+
+* Fixes a few bugs which can lead to node crashes or out-of-memory errors
+
 Kaspad v0.12.2 - 2022-06-17
 ===========================
 

domain/consensus/consensus.go

@@ -194,11 +194,33 @@ func (s *consensus) BuildBlockTemplate(coinbaseData *externalapi.DomainCoinbaseD
 
 // ValidateAndInsertBlock validates the given block and, if valid, applies it
 // to the current state
-func (s *consensus) ValidateAndInsertBlock(block *externalapi.DomainBlock, shouldValidateAgainstUTXO bool) error {
+func (s *consensus) ValidateAndInsertBlock(block *externalapi.DomainBlock, updateVirtual bool) error {
+	if updateVirtual {
+		s.lock.Lock()
+		if s.virtualNotUpdated {
+			s.lock.Unlock()
+			err := s.ResolveVirtual(nil)
+			if err != nil {
+				return err
+			}
+			return s.validateAndInsertBlockWithLock(block, updateVirtual)
+		}
+		defer s.lock.Unlock()
+		_, err := s.validateAndInsertBlockNoLock(block, updateVirtual)
+		if err != nil {
+			return err
+		}
+		return nil
+	}
+
+	return s.validateAndInsertBlockWithLock(block, updateVirtual)
+}
+
+func (s *consensus) validateAndInsertBlockWithLock(block *externalapi.DomainBlock, updateVirtual bool) error {
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	_, err := s.validateAndInsertBlockNoLock(block, shouldValidateAgainstUTXO)
+	_, err := s.validateAndInsertBlockNoLock(block, updateVirtual)
 	if err != nil {
 		return err
 	}
@@ -206,19 +228,6 @@ func (s *consensus) ValidateAndInsertBlock(block *externalapi.DomainBlock, shoul
 }
 
 func (s *consensus) validateAndInsertBlockNoLock(block *externalapi.DomainBlock, updateVirtual bool) (*externalapi.VirtualChangeSet, error) {
-	// If virtual is in non-updated state, and the caller requests updating virtual -- then we must first
-	// resolve virtual so that the new block can be fully processed properly
-	if updateVirtual && s.virtualNotUpdated {
-		for s.virtualNotUpdated {
-			// We use 10000 << finality interval. See comment in `ResolveVirtual`.
-			// We give up responsiveness of consensus in this rare case.
-			_, err := s.resolveVirtualNoLock(10000) // Note `s.virtualNotUpdated` is updated within the call
-			if err != nil {
-				return nil, err
-			}
-		}
-	}
-
 	virtualChangeSet, blockStatus, err := s.blockProcessor.ValidateAndInsertBlock(block, updateVirtual)
 	if err != nil {
 		return nil, err
@@ -257,7 +266,7 @@ func (s *consensus) sendBlockAddedEvent(block *externalapi.DomainBlock, blockSta
 }
 
 func (s *consensus) sendVirtualChangedEvent(virtualChangeSet *externalapi.VirtualChangeSet, wasVirtualUpdated bool) error {
-	if !wasVirtualUpdated || s.consensusEventsChan == nil {
+	if !wasVirtualUpdated || s.consensusEventsChan == nil || virtualChangeSet == nil {
 		return nil
 	}
 
@@ -888,15 +897,42 @@ func (s *consensus) PopulateMass(transaction *externalapi.DomainTransaction) {
 	s.transactionValidator.PopulateMass(transaction)
 }
 
-func (s *consensus) ResolveVirtual() (bool, error) {
+func (s *consensus) ResolveVirtual(progressReportCallback func(uint64, uint64)) error {
+	virtualDAAScoreStart, err := s.GetVirtualDAAScore()
+	if err != nil {
+		return err
+	}
+
+	for i := 0; ; i++ {
+		if i%10 == 0 && progressReportCallback != nil {
+			virtualDAAScore, err := s.GetVirtualDAAScore()
+			if err != nil {
+				return err
+			}
+			progressReportCallback(virtualDAAScoreStart, virtualDAAScore)
+		}
+
+		// In order to prevent a situation that the consensus lock is held for too much time, we
+		// release the lock each time resolve 100 blocks.
+		// Note: maxBlocksToResolve should be smaller than finality interval in order to avoid a situation
+		// where UpdatePruningPointByVirtual skips a pruning point.
+		isCompletelyResolved, err := s.resolveVirtualChunkWithLock(100)
+		if err != nil {
+			return err
+		}
+		if isCompletelyResolved {
+			break
+		}
+	}
+
+	return nil
+}
+
+func (s *consensus) resolveVirtualChunkWithLock(maxBlocksToResolve uint64) (bool, error) {
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	// In order to prevent a situation that the consensus lock is held for too much time, we
-	// release the lock each time resolve 100 blocks.
-	// Note: maxBlocksToResolve should be smaller than finality interval in order to avoid a situation
-	// where UpdatePruningPointByVirtual skips a pruning point.
-	return s.resolveVirtualNoLock(100)
+	return s.resolveVirtualNoLock(maxBlocksToResolve)
 }
 
 func (s *consensus) resolveVirtualNoLock(maxBlocksToResolve uint64) (bool, error) {

domain/consensus/factory.go

@@ -515,6 +515,7 @@ func (f *factory) NewConsensus(config *Config, db infrastructuredatabase.Databas
 		blocksWithTrustedDataDAAWindowStore: daaWindowStore,
 
 		consensusEventsChan: consensusEventsChan,
+		virtualNotUpdated:   true,
 	}
 
 	if isOldReachabilityInitialized {

domain/consensus/finality_test.go

@@ -589,18 +589,13 @@ func TestFinalityResolveVirtual(t *testing.T) {
 			}
 		}
 
-		for i := 0; ; i++ {
-			isCompletelyResolved, err := tc.ResolveVirtual()
-			if err != nil {
-				panic(err)
-			}
-
-			if isCompletelyResolved {
-				t.Log("Resolved virtual")
-				break
-			}
+		err = tc.ResolveVirtual(nil)
+		if err != nil {
+			panic(err)
 		}
 
+		t.Log("Resolved virtual")
+
 		sideChainTipGHOSTDAGData, err = tc.GHOSTDAGDataStore().Get(tc.DatabaseContext(), stagingArea, sideChainTipHash, false)
 		if err != nil {
 			panic(err)

domain/consensus/model/externalapi/blockinfo.go

@@ -13,6 +13,16 @@ type BlockInfo struct {
 	MergeSetReds   []*DomainHash
 }
 
+// HasHeader returns whether the block exists and has a valid header
+func (bi *BlockInfo) HasHeader() bool {
+	return bi.Exists && bi.BlockStatus != StatusInvalid
+}
+
+// HasBody returns whether the block exists and has a valid body
+func (bi *BlockInfo) HasBody() bool {
+	return bi.Exists && bi.BlockStatus != StatusInvalid && bi.BlockStatus != StatusHeaderOnly
+}
+
 // Clone returns a clone of BlockInfo
 func (bi *BlockInfo) Clone() *BlockInfo {
 	return &BlockInfo{

domain/consensus/model/externalapi/consensus.go

@@ -5,7 +5,7 @@ type Consensus interface {
 	Init(skipAddingGenesis bool) error
 	BuildBlock(coinbaseData *DomainCoinbaseData, transactions []*DomainTransaction) (*DomainBlock, error)
 	BuildBlockTemplate(coinbaseData *DomainCoinbaseData, transactions []*DomainTransaction) (*DomainBlockTemplate, error)
-	ValidateAndInsertBlock(block *DomainBlock, shouldValidateAgainstUTXO bool) error
+	ValidateAndInsertBlock(block *DomainBlock, updateVirtual bool) error
 	ValidateAndInsertBlockWithTrustedData(block *BlockWithTrustedData, validateUTXO bool) error
 	ValidateTransactionAndPopulateWithConsensusData(transaction *DomainTransaction) error
 	ImportPruningPoints(pruningPoints []BlockHeader) error
@@ -48,7 +48,7 @@ type Consensus interface {
 	Anticone(blockHash *DomainHash) ([]*DomainHash, error)
 	EstimateNetworkHashesPerSecond(startHash *DomainHash, windowSize int) (uint64, error)
 	PopulateMass(transaction *DomainTransaction)
-	ResolveVirtual() (bool, error)
+	ResolveVirtual(progressReportCallback func(uint64, uint64)) error
 	BlockDAAWindowHashes(blockHash *DomainHash) ([]*DomainHash, error)
 	TrustedDataDataDAAHeader(trustedBlockHash, daaBlockHash *DomainHash, daaBlockWindowIndex uint64) (*TrustedDataDataDAAHeader, error)
 	TrustedBlockAssociatedGHOSTDAGDataBlockHashes(blockHash *DomainHash) ([]*DomainHash, error)

domain/consensus/processes/consensusstatemanager/resolve.go

@@ -5,22 +5,19 @@ import (
 	"github.com/kaspanet/kaspad/domain/consensus/model/externalapi"
 	"github.com/kaspanet/kaspad/infrastructure/logger"
 	"github.com/kaspanet/kaspad/util/staging"
+	"github.com/pkg/errors"
 	"sort"
 )
 
-func (csm *consensusStateManager) ResolveVirtual(maxBlocksToResolve uint64) (*externalapi.VirtualChangeSet, bool, error) {
-	onEnd := logger.LogAndMeasureExecutionTime(log, "csm.ResolveVirtual")
-	defer onEnd()
-
-	readStagingArea := model.NewStagingArea()
-	tips, err := csm.consensusStateStore.Tips(readStagingArea, csm.databaseContext)
+func (csm *consensusStateManager) findNextPendingTip(stagingArea *model.StagingArea) (*externalapi.DomainHash, externalapi.BlockStatus, error) {
+	tips, err := csm.consensusStateStore.Tips(stagingArea, csm.databaseContext)
 	if err != nil {
-		return nil, false, err
+		return nil, externalapi.StatusInvalid, err
 	}
 
 	var sortErr error
 	sort.Slice(tips, func(i, j int) bool {
-		selectedParent, err := csm.ghostdagManager.ChooseSelectedParent(readStagingArea, tips[i], tips[j])
+		selectedParent, err := csm.ghostdagManager.ChooseSelectedParent(stagingArea, tips[i], tips[j])
 		if err != nil {
 			sortErr = err
 			return false
@@ -29,16 +26,14 @@ func (csm *consensusStateManager) ResolveVirtual(maxBlocksToResolve uint64) (*ex
 		return selectedParent.Equal(tips[i])
 	})
 	if sortErr != nil {
-		return nil, false, sortErr
+		return nil, externalapi.StatusInvalid, sortErr
 	}
 
-	var selectedTip *externalapi.DomainHash
-	isCompletelyResolved := true
 	for _, tip := range tips {
 		log.Debugf("Resolving tip %s", tip)
-		isViolatingFinality, shouldNotify, err := csm.isViolatingFinality(readStagingArea, tip)
+		isViolatingFinality, shouldNotify, err := csm.isViolatingFinality(stagingArea, tip)
 		if err != nil {
-			return nil, false, err
+			return nil, externalapi.StatusInvalid, err
 		}
 
 		if isViolatingFinality {
@@ -49,55 +44,143 @@ func (csm *consensusStateManager) ResolveVirtual(maxBlocksToResolve uint64) (*ex
 			continue
 		}
 
-		resolveStagingArea := model.NewStagingArea()
-		unverifiedBlocks, err := csm.getUnverifiedChainBlocks(resolveStagingArea, tip)
+		status, err := csm.blockStatusStore.Get(csm.databaseContext, stagingArea, tip)
 		if err != nil {
-			return nil, false, err
+			return nil, externalapi.StatusInvalid, err
 		}
-
-		resolveTip := tip
-		hasMoreUnverifiedThanMax := maxBlocksToResolve != 0 && uint64(len(unverifiedBlocks)) > maxBlocksToResolve
-		if hasMoreUnverifiedThanMax {
-			resolveTip = unverifiedBlocks[uint64(len(unverifiedBlocks))-maxBlocksToResolve]
-			log.Debugf("Has more than %d blocks to resolve. Changing the resolve tip to %s", maxBlocksToResolve, resolveTip)
-		}
-
-		blockStatus, reversalData, err := csm.resolveBlockStatus(resolveStagingArea, resolveTip, true)
-		if err != nil {
-			return nil, false, err
-		}
-
-		if blockStatus == externalapi.StatusUTXOValid {
-			selectedTip = resolveTip
-			isCompletelyResolved = !hasMoreUnverifiedThanMax
-
-			err = staging.CommitAllChanges(csm.databaseContext, resolveStagingArea)
-			if err != nil {
-				return nil, false, err
-			}
-
-			if reversalData != nil {
-				err = csm.ReverseUTXODiffs(resolveTip, reversalData)
-				if err != nil {
-					return nil, false, err
-				}
-			}
-			break
+		if status == externalapi.StatusUTXOValid || status == externalapi.StatusUTXOPendingVerification {
+			return tip, status, nil
 		}
 	}
 
-	if selectedTip == nil {
-		log.Warnf("Non of the DAG tips are valid")
-		return nil, true, nil
+	return nil, externalapi.StatusInvalid, nil
+}
+
+func (csm *consensusStateManager) getLowerTips(stagingArea *model.StagingArea, pendingTip *externalapi.DomainHash) ([]*externalapi.DomainHash, error) {
+	tips, err := csm.consensusStateStore.Tips(stagingArea, csm.databaseContext)
+	if err != nil {
+		return nil, err
 	}
 
-	oldVirtualGHOSTDAGData, err := csm.ghostdagDataStore.Get(csm.databaseContext, readStagingArea, model.VirtualBlockHash, false)
+	lowerTips := []*externalapi.DomainHash{pendingTip}
+	for _, tip := range tips {
+		if tip.Equal(pendingTip) {
+			continue
+		}
+		selectedParent, err := csm.ghostdagManager.ChooseSelectedParent(stagingArea, tip, pendingTip)
+		if err != nil {
+			return nil, err
+		}
+		if selectedParent.Equal(pendingTip) {
+			lowerTips = append(lowerTips, tip)
+		}
+	}
+	return lowerTips, nil
+}
+
+func (csm *consensusStateManager) ResolveVirtual(maxBlocksToResolve uint64) (*externalapi.VirtualChangeSet, bool, error) {
+	onEnd := logger.LogAndMeasureExecutionTime(log, "csm.ResolveVirtual")
+	defer onEnd()
+
+	readStagingArea := model.NewStagingArea()
+
+	pendingTip, pendingTipStatus, err := csm.findNextPendingTip(readStagingArea)
 	if err != nil {
 		return nil, false, err
 	}
 
+	if pendingTip == nil {
+		log.Warnf("Non of the DAG tips are valid")
+		return nil, true, nil
+	}
+
+	prevVirtualSelectedParent, err := csm.virtualSelectedParent(readStagingArea)
+	if err != nil {
+		return nil, false, err
+	}
+
+	if pendingTipStatus == externalapi.StatusUTXOValid && prevVirtualSelectedParent.Equal(pendingTip) {
+		return nil, true, nil
+	}
+
+	// Resolve a chunk from the pending chain
+	resolveStagingArea := model.NewStagingArea()
+	unverifiedBlocks, err := csm.getUnverifiedChainBlocks(resolveStagingArea, pendingTip)
+	if err != nil {
+		return nil, false, err
+	}
+
+	intermediateTip := pendingTip
+
+	// Too many blocks to verify, so we only process a chunk and return
+	if maxBlocksToResolve != 0 && uint64(len(unverifiedBlocks)) > maxBlocksToResolve {
+
+		intermediateTipIndex := uint64(len(unverifiedBlocks)) - maxBlocksToResolve
+		intermediateTip = unverifiedBlocks[intermediateTipIndex]
+		isNewVirtualSelectedParent, err := csm.isNewSelectedTip(readStagingArea, intermediateTip, prevVirtualSelectedParent)
+		if err != nil {
+			return nil, false, err
+		}
+
+		// We must find an intermediate tip which wins previous virtual selected parent
+		// even if we process more than `maxBlocksToResolve` for that.
+		// Otherwise, internal UTXO diff logic gets all messed up
+		for !isNewVirtualSelectedParent {
+			if intermediateTipIndex == 0 {
+				return nil, false, errors.Errorf(
+					"Expecting the pending tip %s to overcome the previous selected parent %s", pendingTip, prevVirtualSelectedParent)
+			}
+			intermediateTipIndex--
+			intermediateTip = unverifiedBlocks[intermediateTipIndex]
+			isNewVirtualSelectedParent, err = csm.isNewSelectedTip(readStagingArea, intermediateTip, prevVirtualSelectedParent)
+			if err != nil {
+				return nil, false, err
+			}
+		}
+		log.Debugf("Has more than %d blocks to resolve. Changing the resolve tip to %s", maxBlocksToResolve, intermediateTip)
+	}
+
+	intermediateTipStatus, reversalData, err := csm.resolveBlockStatus(
+		resolveStagingArea, intermediateTip, true)
+	if err != nil {
+		return nil, false, err
+	}
+
+	if intermediateTipStatus == externalapi.StatusUTXOValid {
+		err = staging.CommitAllChanges(csm.databaseContext, resolveStagingArea)
+		if err != nil {
+			return nil, false, err
+		}
+
+		if reversalData != nil {
+			err = csm.ReverseUTXODiffs(intermediateTip, reversalData)
+			if err != nil {
+				return nil, false, err
+			}
+		}
+	}
+
+	isActualTip := intermediateTip.Equal(pendingTip)
+	isCompletelyResolved := isActualTip && intermediateTipStatus == externalapi.StatusUTXOValid
+
 	updateVirtualStagingArea := model.NewStagingArea()
-	virtualUTXODiff, err := csm.updateVirtualWithParents(updateVirtualStagingArea, []*externalapi.DomainHash{selectedTip})
+
+	// If `isCompletelyResolved`, set virtual correctly with all tips which have less blue work than pending
+	virtualTipCandidates := []*externalapi.DomainHash{intermediateTip}
+	if isCompletelyResolved {
+		lowerTips, err := csm.getLowerTips(readStagingArea, pendingTip)
+		if err != nil {
+			return nil, false, err
+		}
+
+		log.Debugf("Picking virtual parents from relevant tips len: %d", len(lowerTips))
+		virtualTipCandidates, err = csm.pickVirtualParents(readStagingArea, lowerTips)
+		if err != nil {
+			return nil, false, err
+		}
+		log.Debugf("Picked virtual parents: %s", virtualTipCandidates)
+	}
+	virtualUTXODiff, err := csm.updateVirtualWithParents(updateVirtualStagingArea, virtualTipCandidates)
 	if err != nil {
 		return nil, false, err
 	}
@@ -108,12 +191,12 @@ func (csm *consensusStateManager) ResolveVirtual(maxBlocksToResolve uint64) (*ex
 	}
 
 	selectedParentChainChanges, err := csm.dagTraversalManager.
-		CalculateChainPath(readStagingArea, oldVirtualGHOSTDAGData.SelectedParent(), selectedTip)
+		CalculateChainPath(updateVirtualStagingArea, prevVirtualSelectedParent, pendingTip)
 	if err != nil {
 		return nil, false, err
 	}
 
-	virtualParents, err := csm.dagTopologyManager.Parents(readStagingArea, model.VirtualBlockHash)
+	virtualParents, err := csm.dagTopologyManager.Parents(updateVirtualStagingArea, model.VirtualBlockHash)
 	if err != nil {
 		return nil, false, err
 	}

domain/consensus/processes/consensusstatemanager/resolve_block_status.go

@@ -233,7 +233,7 @@ func (csm *consensusStateManager) resolveSingleBlockStatus(stagingArea *model.St
 		return externalapi.StatusUTXOValid, nil, nil
 	}
 
-	oldSelectedTip, err := csm.selectedTip(stagingArea)
+	oldSelectedTip, err := csm.virtualSelectedParent(stagingArea)
 	if err != nil {
 		return 0, nil, err
 	}
@@ -298,7 +298,7 @@ func (csm *consensusStateManager) isNewSelectedTip(stagingArea *model.StagingAre
 	return blockHash.Equal(newSelectedTip), nil
 }
 
-func (csm *consensusStateManager) selectedTip(stagingArea *model.StagingArea) (*externalapi.DomainHash, error) {
+func (csm *consensusStateManager) virtualSelectedParent(stagingArea *model.StagingArea) (*externalapi.DomainHash, error) {
 	virtualGHOSTDAGData, err := csm.ghostdagDataStore.Get(csm.databaseContext, stagingArea, model.VirtualBlockHash, false)
 	if err != nil {
 		return nil, err

domain/consensus/processes/consensusstatemanager/resolve_virtual_test.go

@@ -1,6 +1,9 @@
 package consensusstatemanager_test
 
 import (
+	"fmt"
+	"github.com/kaspanet/kaspad/domain/consensus/model"
+	"github.com/kaspanet/kaspad/domain/consensus/model/testapi"
 	"github.com/kaspanet/kaspad/domain/consensus/utils/consensushashing"
 	"testing"
 
@@ -212,3 +215,106 @@ func TestAddGenesisChildAfterTwoResolveVirtualCalls(t *testing.T) {
 		}
 	})
 }
+
+func TestResolveVirtualMess(t *testing.T) {
+
+	testutils.ForAllNets(t, true, func(t *testing.T, consensusConfig *consensus.Config) {
+		factory := consensus.NewFactory()
+
+		tc, teardown, err := factory.NewTestConsensus(consensusConfig, "TestAddGenesisChildAfterTwoResolveVirtualCalls")
+		if err != nil {
+			t.Fatalf("Error setting up consensus: %+v", err)
+		}
+		defer teardown(false)
+
+		hashes := []*externalapi.DomainHash{consensusConfig.GenesisHash}
+		blocks := make(map[externalapi.DomainHash]string)
+		blocks[*consensusConfig.GenesisHash] = "g"
+		blocks[*model.VirtualBlockHash] = "v"
+		//fmt.Printf("%s\n\n", consensusConfig.GenesisHash)
+
+		// Create a chain of blocks
+		const initialChainLength = 6
+		previousBlockHash := consensusConfig.GenesisHash
+		for i := 0; i < initialChainLength; i++ {
+			previousBlockHash, _, err = tc.AddBlock([]*externalapi.DomainHash{previousBlockHash}, nil, nil)
+			blocks[*previousBlockHash] = fmt.Sprintf("A_%d", i)
+			hashes = append(hashes, previousBlockHash)
+			//fmt.Printf("A_%d: %s\n", i, previousBlockHash)
+
+			if err != nil {
+				t.Fatalf("Error mining block no. %d in initial chain: %+v", i, err)
+			}
+		}
+
+		//fmt.Printf("\n")
+
+		// Mine a chain with more blocks, to re-organize the DAG
+		const reorgChainLength = 20 // initialChainLength + 1
+		previousBlockHash = consensusConfig.GenesisHash
+		for i := 0; i < reorgChainLength; i++ {
+			previousBlock, _, err := tc.BuildBlockWithParents([]*externalapi.DomainHash{previousBlockHash}, nil, nil)
+			if err != nil {
+				t.Fatalf("Error mining block no. %d in re-org chain: %+v", i, err)
+			}
+			previousBlockHash = consensushashing.BlockHash(previousBlock)
+			blocks[*previousBlockHash] = fmt.Sprintf("B_%d", i)
+			hashes = append(hashes, previousBlockHash)
+			//fmt.Printf("B_%d: %s\n", i, previousBlockHash)
+
+			// Do not UTXO validate in order to resolve virtual later
+			err = tc.ValidateAndInsertBlock(previousBlock, false)
+			if err != nil {
+				t.Fatalf("Error mining block no. %d in re-org chain: %+v", i, err)
+			}
+		}
+
+		//fmt.Printf("\n")
+
+		//printUtxoDiffChildren(t, hashes, tc, blocks)
+
+		// Resolve one step
+		_, err = tc.ResolveVirtualWithMaxParam(3)
+		if err != nil {
+			printUtxoDiffChildren(t, hashes, tc, blocks)
+			t.Fatalf("Error resolving virtual in re-org chain: %+v", err)
+		}
+
+		// Resolve one more step
+		isCompletelyResolved, err := tc.ResolveVirtualWithMaxParam(3)
+		if err != nil {
+			t.Fatalf("Error resolving virtual in re-org chain: %+v", err)
+		}
+
+		// Complete resolving virtual
+		for !isCompletelyResolved {
+			isCompletelyResolved, err = tc.ResolveVirtualWithMaxParam(3)
+			if err != nil {
+				t.Fatalf("Error resolving virtual in re-org chain: %+v", err)
+			}
+		}
+
+		//printUtxoDiffChildren(t, hashes, tc, blocks)
+	})
+}
+
+func printUtxoDiffChildren(t *testing.T, hashes []*externalapi.DomainHash, tc testapi.TestConsensus, blocks map[externalapi.DomainHash]string) {
+	fmt.Printf("\n===============================\nBlock\t\tDiff child\n")
+	stagingArea := model.NewStagingArea()
+	for _, block := range hashes {
+		hasUTXODiffChild, err := tc.UTXODiffStore().HasUTXODiffChild(tc.DatabaseContext(), stagingArea, block)
+		if err != nil {
+			t.Fatalf("Error while reading utxo diff store: %+v", err)
+		}
+		if hasUTXODiffChild {
+			utxoDiffChild, err := tc.UTXODiffStore().UTXODiffChild(tc.DatabaseContext(), stagingArea, block)
+			if err != nil {
+				t.Fatalf("Error while reading utxo diff store: %+v", err)
+			}
+			fmt.Printf("%s\t\t\t%s\n", blocks[*block], blocks[*utxoDiffChild])
+		} else {
+			fmt.Printf("%s\n", blocks[*block])
+		}
+	}
+	fmt.Printf("\n===============================\n")
+}

domain/consensus/processes/consensusstatemanager/reverse_utxo_diffs.go

@@ -56,12 +56,6 @@ func (csm *consensusStateManager) ReverseUTXODiffs(tipHash *externalapi.DomainHa
 			return err
 		}
 
-		// We stop reversing when current's UTXODiffChild is not current's SelectedParent
-		if !currentBlockGHOSTDAGData.SelectedParent().Equal(currentBlockUTXODiffChild) {
-			log.Debugf("Block %s's UTXODiffChild is not it's selected parent - finish reversing", currentBlock)
-			break
-		}
-
 		currentUTXODiff := previousUTXODiff.Reversed()
 
 		// retrieve current utxoDiff for Bi, to be used by next block
@@ -75,6 +69,12 @@ func (csm *consensusStateManager) ReverseUTXODiffs(tipHash *externalapi.DomainHa
 			return err
 		}
 
+		// We stop reversing when current's UTXODiffChild is not current's SelectedParent
+		if !currentBlockGHOSTDAGData.SelectedParent().Equal(currentBlockUTXODiffChild) {
+			log.Debugf("Block %s's UTXODiffChild is not it's selected parent - finish reversing", currentBlock)
+			break
+		}
+
 		previousBlock = currentBlock
 		previousBlockGHOSTDAGData = currentBlockGHOSTDAGData
 

domain/consensus/processes/consensusstatemanager/update_virtual.go

@@ -110,7 +110,7 @@ func (csm *consensusStateManager) updateSelectedTipUTXODiff(
 	onEnd := logger.LogAndMeasureExecutionTime(log, "updateSelectedTipUTXODiff")
 	defer onEnd()
 
-	selectedTip, err := csm.selectedTip(stagingArea)
+	selectedTip, err := csm.virtualSelectedParent(stagingArea)
 	if err != nil {
 		return err
 	}

domain/migrate.go

@@ -214,34 +214,19 @@ func syncConsensuses(syncer, syncee externalapi.Consensus) error {
 		return err
 	}
 
-	virtualDAAScoreStart, err := syncee.GetVirtualDAAScore()
+	err = syncer.ResolveVirtual(func(virtualDAAScoreStart uint64, virtualDAAScore uint64) {
+		if estimatedVirtualDAAScoreTarget-virtualDAAScoreStart <= 0 {
+			percents = 100
+		} else {
+			percents = int(float64(virtualDAAScore-virtualDAAScoreStart) / float64(estimatedVirtualDAAScoreTarget-virtualDAAScoreStart) * 100)
+		}
+		log.Infof("Resolving virtual. Estimated progress: %d%%", percents)
+	})
 	if err != nil {
 		return err
 	}
 
-	percents = 0
-	for i := 0; ; i++ {
-		if i%10 == 0 {
-			virtualDAAScore, err := syncee.GetVirtualDAAScore()
-			if err != nil {
-				return err
-			}
-			newPercents := int(float64(virtualDAAScore-virtualDAAScoreStart) / float64(estimatedVirtualDAAScoreTarget-virtualDAAScoreStart) * 100)
-			if newPercents > percents {
-				percents = newPercents
-				log.Infof("Resolving virtual. Estimated progress: %d%%", percents)
-			}
-		}
-		isCompletelyResolved, err := syncee.ResolveVirtual()
-		if err != nil {
-			return err
-		}
-
-		if isCompletelyResolved {
-			log.Infof("Resolved virtual")
-			break
-		}
-	}
+	log.Infof("Resolved virtual")
 
 	return nil
 }

infrastructure/config/config.go

@@ -98,6 +98,7 @@ type Flags struct {
 	RPCMaxWebsockets                int           `long:"rpcmaxwebsockets" description:"Max number of RPC websocket connections"`
 	RPCMaxConcurrentReqs            int           `long:"rpcmaxconcurrentreqs" description:"Max number of concurrent RPC requests that may be processed concurrently"`
 	DisableRPC                      bool          `long:"norpc" description:"Disable built-in RPC server"`
+	SafeRPC                         bool          `long:"saferpc" description:"Disable RPC commands which affect the state of the node"`
 	DisableDNSSeed                  bool          `long:"nodnsseed" description:"Disable DNS seeding for peers"`
 	DNSSeed                         string        `long:"dnsseed" description:"Override DNS seeds with specified hostname (Only 1 hostname allowed)"`
 	GRPCSeed                        string        `long:"grpcseed" description:"Hostname of gRPC server for seeding peers"`

version/version.go

@@ -11,7 +11,7 @@ const validCharacters = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrs
 const (
 	appMajor uint = 0
 	appMinor uint = 12
-	appPatch uint = 2
+	appPatch uint = 4
 )
 
 // appBuild is defined as a variable so it can be overridden during the build