Merge branch 'dev' into replace-daglabs-dnsseeder

app/rpc/manager.go

@@ -223,18 +223,9 @@ func (m *Manager) notifyVirtualSelectedParentChainChanged(virtualChangeSet *exte
 	onEnd := logger.LogAndMeasureExecutionTime(log, "RPCManager.NotifyVirtualSelectedParentChainChanged")
 	defer onEnd()
 
-	listenersThatPropagateSelectedParentChanged :=
-		m.context.NotificationManager.AllListenersThatPropagateVirtualSelectedParentChainChanged()
-	if len(listenersThatPropagateSelectedParentChanged) > 0 {
-		// Generating acceptedTransactionIDs is a heavy operation, so we check if it's needed by any listener.
-		includeAcceptedTransactionIDs := false
-		for _, listener := range listenersThatPropagateSelectedParentChanged {
-			if listener.IncludeAcceptedTransactionIDsInVirtualSelectedParentChainChangedNotifications() {
-				includeAcceptedTransactionIDs = true
-				break
-			}
-		}
+	hasListeners, includeAcceptedTransactionIDs := m.context.NotificationManager.HasListenersThatPropagateVirtualSelectedParentChainChanged()
 
+	if hasListeners {
 		notification, err := m.context.ConvertVirtualSelectedParentChainChangesToChainChangedNotificationMessage(
 			virtualChangeSet.VirtualSelectedParentChainChanges, includeAcceptedTransactionIDs)
 		if err != nil {

app/rpc/rpccontext/notificationmanager.go

@@ -142,16 +142,29 @@ func (nm *NotificationManager) NotifyVirtualSelectedParentChainChanged(
 	return nil
 }
 
-// AllListenersThatPropagateVirtualSelectedParentChainChanged returns true if there's any listener that is
-// subscribed to VirtualSelectedParentChainChanged notifications.
-func (nm *NotificationManager) AllListenersThatPropagateVirtualSelectedParentChainChanged() []*NotificationListener {
-	var listenersThatPropagate []*NotificationListener
+// HasListenersThatPropagateVirtualSelectedParentChainChanged returns whether there's any listener that is
+// subscribed to VirtualSelectedParentChainChanged notifications as well as checks if any such listener requested
+// to include AcceptedTransactionIDs.
+func (nm *NotificationManager) HasListenersThatPropagateVirtualSelectedParentChainChanged() (hasListeners, hasListenersThatRequireAcceptedTransactionIDs bool) {
+
+	nm.RLock()
+	defer nm.RUnlock()
+
+	hasListeners = false
+	hasListenersThatRequireAcceptedTransactionIDs = false
+
 	for _, listener := range nm.listeners {
 		if listener.propagateVirtualSelectedParentChainChangedNotifications {
-			listenersThatPropagate = append(listenersThatPropagate, listener)
+			hasListeners = true
+			// Generating acceptedTransactionIDs is a heavy operation, so we check if it's needed by any listener.
+			if listener.includeAcceptedTransactionIDsInVirtualSelectedParentChainChangedNotifications {
+				hasListenersThatRequireAcceptedTransactionIDs = true
+				break
 			}
 		}
-	return listenersThatPropagate
+	}
+
+	return hasListeners, hasListenersThatRequireAcceptedTransactionIDs
 }
 
 // NotifyFinalityConflict notifies the notification manager that there's a finality conflict in the DAG
@@ -338,7 +351,11 @@ func (nl *NotificationListener) PropagateFinalityConflictResolvedNotifications()
 // to the remote listener for the given addresses. Subsequent calls instruct the listener to
 // send UTXOs changed notifications for those addresses along with the old ones. Duplicate addresses
 // are ignored.
-func (nl *NotificationListener) PropagateUTXOsChangedNotifications(addresses []*UTXOsChangedNotificationAddress) {
+func (nm *NotificationManager) PropagateUTXOsChangedNotifications(nl *NotificationListener, addresses []*UTXOsChangedNotificationAddress) {
+	// Apply a write-lock since the internal listener address map is modified
+	nm.Lock()
+	defer nm.Unlock()
+
 	if !nl.propagateUTXOsChangedNotifications {
 		nl.propagateUTXOsChangedNotifications = true
 		nl.propagateUTXOsChangedNotificationAddresses =
@@ -353,7 +370,11 @@ func (nl *NotificationListener) PropagateUTXOsChangedNotifications(addresses []*
 // StopPropagatingUTXOsChangedNotifications instructs the listener to stop sending UTXOs
 // changed notifications to the remote listener for the given addresses. Addresses for which
 // notifications are not currently sent are ignored.
-func (nl *NotificationListener) StopPropagatingUTXOsChangedNotifications(addresses []*UTXOsChangedNotificationAddress) {
+func (nm *NotificationManager) StopPropagatingUTXOsChangedNotifications(nl *NotificationListener, addresses []*UTXOsChangedNotificationAddress) {
+	// Apply a write-lock since the internal listener address map is modified
+	nm.Lock()
+	defer nm.Unlock()
+
 	if !nl.propagateUTXOsChangedNotifications {
 		return
 	}

app/rpc/rpchandlers/notify_utxos_changed.go

@@ -26,7 +26,7 @@ func HandleNotifyUTXOsChanged(context *rpccontext.Context, router *router.Router
 	if err != nil {
 		return nil, err
 	}
-	listener.PropagateUTXOsChangedNotifications(addresses)
+	context.NotificationManager.PropagateUTXOsChangedNotifications(listener, addresses)
 
 	response := appmessage.NewNotifyUTXOsChangedResponseMessage()
 	return response, nil

app/rpc/rpchandlers/stop_notifying_utxos_changed.go

@@ -26,7 +26,7 @@ func HandleStopNotifyingUTXOsChanged(context *rpccontext.Context, router *router
 	if err != nil {
 		return nil, err
 	}
-	listener.StopPropagatingUTXOsChangedNotifications(addresses)
+	context.NotificationManager.StopPropagatingUTXOsChangedNotifications(listener, addresses)
 
 	response := appmessage.NewStopNotifyingUTXOsChangedResponseMessage()
 	return response, nil

domain/consensus/consensus.go

@@ -64,6 +64,12 @@ type consensus struct {
 	virtualNotUpdated   bool
 }
 
+// In order to prevent a situation that the consensus lock is held for too much time, we
+// release the lock each time we resolve 100 blocks.
+// Note: `virtualResolveChunk` should be smaller than `params.FinalityDuration` in order to avoid a situation
+// where UpdatePruningPointByVirtual skips a pruning point.
+const virtualResolveChunk = 100
+
 func (s *consensus) ValidateAndInsertBlockWithTrustedData(block *externalapi.BlockWithTrustedData, validateUTXO bool) error {
 	s.lock.Lock()
 	defer s.lock.Unlock()
@@ -198,15 +204,32 @@ func (s *consensus) ValidateAndInsertBlock(block *externalapi.DomainBlock, updat
 	if updateVirtual {
 		s.lock.Lock()
 		if s.virtualNotUpdated {
+			// We enter the loop in locked state
+			for {
+				_, isCompletelyResolved, err := s.resolveVirtualChunkNoLock(virtualResolveChunk)
+				if err != nil {
+					s.lock.Unlock()
+					return err
+				}
+				if isCompletelyResolved {
+					// Make sure we enter the block insertion function w/o releasing the lock.
+					// Otherwise, we might actually enter it in `s.virtualNotUpdated == true` state
+					_, err = s.validateAndInsertBlockNoLock(block, updateVirtual)
+					// Finally, unlock for the last iteration and return
 					s.lock.Unlock()
-			err := s.ResolveVirtual(nil)
 					if err != nil {
 						return err
 					}
-			return s.validateAndInsertBlockWithLock(block, updateVirtual)
+					return nil
+				}
+				// Unlock to allow other threads to enter consensus
+				s.lock.Unlock()
+				// Lock for the next iteration
+				s.lock.Lock()
+			}
 		}
-		defer s.lock.Unlock()
 		_, err := s.validateAndInsertBlockNoLock(block, updateVirtual)
+		s.lock.Unlock()
 		if err != nil {
 			return err
 		}
@@ -912,11 +935,7 @@ func (s *consensus) ResolveVirtual(progressReportCallback func(uint64, uint64))
 			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 we resolve 100 blocks.
-		// Note: maxBlocksToResolve should be smaller than `params.FinalityDuration` in order to avoid a situation
-		// where UpdatePruningPointByVirtual skips a pruning point.
-		_, isCompletelyResolved, err := s.resolveVirtualChunkWithLock(100)
+		_, isCompletelyResolved, err := s.resolveVirtualChunkWithLock(virtualResolveChunk)
 		if err != nil {
 			return err
 		}

domain/consensus/processes/pruningmanager/pruningmanager.go

@@ -772,78 +772,41 @@ func (pm *pruningManager) calculateDiffBetweenPreviousAndCurrentPruningPoints(st
 	if err != nil {
 		return nil, err
 	}
-	currentPruningGhostDAG, err := pm.ghostdagDataStore.Get(pm.databaseContext, stagingArea, currentPruningHash, false)
+
+	utxoDiff := utxo.NewMutableUTXODiff()
+
+	iterator, err := pm.dagTraversalManager.SelectedChildIterator(stagingArea, currentPruningHash, previousPruningHash, false)
 	if err != nil {
 		return nil, err
 	}
-	previousPruningGhostDAG, err := pm.ghostdagDataStore.Get(pm.databaseContext, stagingArea, previousPruningHash, false)
+	defer iterator.Close()
+
+	for ok := iterator.First(); ok; ok = iterator.Next() {
+		child, err := iterator.Get()
 		if err != nil {
 			return nil, err
 		}
+		chainBlockAcceptanceData, err := pm.acceptanceDataStore.Get(pm.databaseContext, stagingArea, child)
+		if err != nil {
+			return nil, err
+		}
+		chainBlockHeader, err := pm.blockHeaderStore.BlockHeader(pm.databaseContext, stagingArea, child)
+		if err != nil {
+			return nil, err
+		}
+		for _, blockAcceptanceData := range chainBlockAcceptanceData {
+			for _, transactionAcceptanceData := range blockAcceptanceData.TransactionAcceptanceData {
+				if transactionAcceptanceData.IsAccepted {
+					err = utxoDiff.AddTransaction(transactionAcceptanceData.Transaction, chainBlockHeader.DAAScore())
+					if err != nil {
+						return nil, err
+					}
+				}
+			}
+		}
+	}
 
-	currentPruningCurrentDiffChild := currentPruningHash
-	previousPruningCurrentDiffChild := previousPruningHash
-	// We need to use BlueWork because it's the only thing that's monotonic in the whole DAG
-	// We use the BlueWork to know which point is currently lower on the DAG so we can keep climbing its children,
-	// that way we keep climbing on the lowest point until they both reach the exact same descendant
-	currentPruningCurrentDiffChildBlueWork := currentPruningGhostDAG.BlueWork()
-	previousPruningCurrentDiffChildBlueWork := previousPruningGhostDAG.BlueWork()
-
-	var diffHashesFromPrevious []*externalapi.DomainHash
-	var diffHashesFromCurrent []*externalapi.DomainHash
-	for {
-		// if currentPruningCurrentDiffChildBlueWork > previousPruningCurrentDiffChildBlueWork
-		if currentPruningCurrentDiffChildBlueWork.Cmp(previousPruningCurrentDiffChildBlueWork) == 1 {
-			diffHashesFromPrevious = append(diffHashesFromPrevious, previousPruningCurrentDiffChild)
-			previousPruningCurrentDiffChild, err = pm.utxoDiffStore.UTXODiffChild(pm.databaseContext, stagingArea, previousPruningCurrentDiffChild)
-			if err != nil {
-				return nil, err
-			}
-			diffChildGhostDag, err := pm.ghostdagDataStore.Get(pm.databaseContext, stagingArea, previousPruningCurrentDiffChild, false)
-			if err != nil {
-				return nil, err
-			}
-			previousPruningCurrentDiffChildBlueWork = diffChildGhostDag.BlueWork()
-		} else if currentPruningCurrentDiffChild.Equal(previousPruningCurrentDiffChild) {
-			break
-		} else {
-			diffHashesFromCurrent = append(diffHashesFromCurrent, currentPruningCurrentDiffChild)
-			currentPruningCurrentDiffChild, err = pm.utxoDiffStore.UTXODiffChild(pm.databaseContext, stagingArea, currentPruningCurrentDiffChild)
-			if err != nil {
-				return nil, err
-			}
-			diffChildGhostDag, err := pm.ghostdagDataStore.Get(pm.databaseContext, stagingArea, currentPruningCurrentDiffChild, false)
-			if err != nil {
-				return nil, err
-			}
-			currentPruningCurrentDiffChildBlueWork = diffChildGhostDag.BlueWork()
-		}
-	}
-	// The order in which we apply the diffs should be from top to bottom, but we traversed from bottom to top
-	// so we apply the diffs in reverse order.
-	oldDiff := utxo.NewMutableUTXODiff()
-	for i := len(diffHashesFromPrevious) - 1; i >= 0; i-- {
-		utxoDiff, err := pm.utxoDiffStore.UTXODiff(pm.databaseContext, stagingArea, diffHashesFromPrevious[i])
-		if err != nil {
-			return nil, err
-		}
-		err = oldDiff.WithDiffInPlace(utxoDiff)
-		if err != nil {
-			return nil, err
-		}
-	}
-	newDiff := utxo.NewMutableUTXODiff()
-	for i := len(diffHashesFromCurrent) - 1; i >= 0; i-- {
-		utxoDiff, err := pm.utxoDiffStore.UTXODiff(pm.databaseContext, stagingArea, diffHashesFromCurrent[i])
-		if err != nil {
-			return nil, err
-		}
-		err = newDiff.WithDiffInPlace(utxoDiff)
-		if err != nil {
-			return nil, err
-		}
-	}
-	return oldDiff.DiffFrom(newDiff.ToImmutable())
+	return utxoDiff.ToImmutable(), err
 }
 
 // finalityScore is the number of finality intervals passed since

domain/dagconfig/params.go

@@ -297,7 +297,7 @@ var TestnetParams = Params{
 	Net:         appmessage.Testnet,
 	RPCPort:     "16210",
 	DefaultPort: "16211",
-	DNSSeeds:    []string{"testnet-9-dnsseed.daglabs-dev.com"},
+	DNSSeeds:    []string{"testnet-10-dnsseed.kas.pa"},
 
 	// DAG parameters
 	GenesisBlock:                    &testnetGenesisBlock,

infrastructure/network/rpcclient/grpcclient/grpcclient.go

@@ -22,6 +22,7 @@ type OnDisconnectedHandler func()
 // GRPCClient is a gRPC-based RPC client
 type GRPCClient struct {
 	stream                protowire.RPC_MessageStreamClient
+	connection            *grpc.ClientConn
 	onErrorHandler        OnErrorHandler
 	onDisconnectedHandler OnDisconnectedHandler
 }
@@ -43,7 +44,12 @@ func Connect(address string) (*GRPCClient, error) {
 	if err != nil {
 		return nil, errors.Wrapf(err, "error getting client stream for %s", address)
 	}
-	return &GRPCClient{stream: stream}, nil
+	return &GRPCClient{stream: stream, connection: gRPCConnection}, nil
+}
+
+// Close closes the underlying grpc connection
+func (c *GRPCClient) Close() error {
+	return c.connection.Close()
 }
 
 // Disconnect disconnects from the RPC server

infrastructure/network/rpcclient/rpcclient.go

@@ -143,6 +143,9 @@ func (c *RPCClient) handleClientDisconnected() {
 }
 
 func (c *RPCClient) handleClientError(err error) {
+	if atomic.LoadUint32(&c.isClosed) == 1 {
+		return
+	}
 	log.Warnf("Received error from client: %s", err)
 	c.handleClientDisconnected()
 }
@@ -159,7 +162,7 @@ func (c *RPCClient) Close() error {
 		return errors.Errorf("Cannot close a client that had already been closed")
 	}
 	c.rpcRouter.router.Close()
-	return nil
+	return c.GRPCClient.Close()
 }
 
 // Address returns the address the RPC client connected to

testing/integration/rpc_test.go

@@ -2,6 +2,7 @@ package integration
 
 import (
 	"github.com/kaspanet/kaspad/infrastructure/config"
+	"runtime"
 	"testing"
 	"time"
 
@@ -26,6 +27,37 @@ func newTestRPCClient(rpcAddress string) (*testRPCClient, error) {
 	}, nil
 }
 
+func connectAndClose(rpcAddress string) error {
+	client, err := rpcclient.NewRPCClient(rpcAddress)
+	if err != nil {
+		return err
+	}
+	defer client.Close()
+	return nil
+}
+
+func TestRPCClientGoroutineLeak(t *testing.T) {
+	_, teardown := setupHarness(t, &harnessParams{
+		p2pAddress:              p2pAddress1,
+		rpcAddress:              rpcAddress1,
+		miningAddress:           miningAddress1,
+		miningAddressPrivateKey: miningAddress1PrivateKey,
+	})
+	defer teardown()
+	numGoroutinesBefore := runtime.NumGoroutine()
+	for i := 1; i < 100; i++ {
+		err := connectAndClose(rpcAddress1)
+		if err != nil {
+			t.Fatalf("Failed to set up an RPC client: %s", err)
+		}
+		time.Sleep(10 * time.Millisecond)
+		if runtime.NumGoroutine() > numGoroutinesBefore+10 {
+			t.Fatalf("Number of goroutines is increasing for each RPC client open (%d -> %d), which indicates a memory leak",
+				numGoroutinesBefore, runtime.NumGoroutine())
+		}
+	}
+}
+
 func TestRPCMaxInboundConnections(t *testing.T) {
 	harness, teardown := setupHarness(t, &harnessParams{
 		p2pAddress:              p2pAddress1,