core, les, eth: port snap sync changes (#24898)

core, eth, les, trie: rework snap sync

core/state/statedb.go

@@ -940,7 +940,7 @@ func (s *StateDB) Commit(deleteEmptyObjects bool) (common.Hash, error) {
 	// The onleaf func is called _serially_, so we can reuse the same account
 	// for unmarshalling every time.
 	var account types.StateAccount
-	root, accountCommitted, err := s.trie.Commit(func(_ [][]byte, _ []byte, leaf []byte, parent common.Hash) error {
+	root, accountCommitted, err := s.trie.Commit(func(_ [][]byte, _ []byte, leaf []byte, parent common.Hash, _ []byte) error {
 		if err := rlp.DecodeBytes(leaf, &account); err != nil {
 			return nil
 		}

core/state/sync.go

@@ -27,20 +27,20 @@ import (
 )
 
 // NewStateSync create a new state trie download scheduler.
-func NewStateSync(root common.Hash, database ethdb.KeyValueReader, onLeaf func(paths [][]byte, leaf []byte) error) *trie.Sync {
+func NewStateSync(root common.Hash, database ethdb.KeyValueReader, onLeaf func(keys [][]byte, leaf []byte) error) *trie.Sync {
 	// Register the storage slot callback if the external callback is specified.
-	var onSlot func(paths [][]byte, hexpath []byte, leaf []byte, parent common.Hash) error
+	var onSlot func(keys [][]byte, path []byte, leaf []byte, parent common.Hash, parentPath []byte) error
 	if onLeaf != nil {
-		onSlot = func(paths [][]byte, hexpath []byte, leaf []byte, parent common.Hash) error {
-			return onLeaf(paths, leaf)
+		onSlot = func(keys [][]byte, path []byte, leaf []byte, parent common.Hash, parentPath []byte) error {
+			return onLeaf(keys, leaf)
 		}
 	}
 	// Register the account callback to connect the state trie and the storage
 	// trie belongs to the contract.
 	var syncer *trie.Sync
-	onAccount := func(paths [][]byte, hexpath []byte, leaf []byte, parent common.Hash) error {
+	onAccount := func(keys [][]byte, path []byte, leaf []byte, parent common.Hash, parentPath []byte) error {
 		if onLeaf != nil {
-			if err := onLeaf(paths, leaf); err != nil {
+			if err := onLeaf(keys, leaf); err != nil {
 				return err
 			}
 		}
@@ -48,8 +48,8 @@ func NewStateSync(root common.Hash, database ethdb.KeyValueReader, onLeaf func(p
 		if err := rlp.Decode(bytes.NewReader(leaf), &obj); err != nil {
 			return err
 		}
-		syncer.AddSubTrie(obj.Root, hexpath, parent, onSlot)
-		syncer.AddCodeEntry(common.BytesToHash(obj.CodeHash), hexpath, parent)
+		syncer.AddSubTrie(obj.Root, path, parent, parentPath, onSlot)
+		syncer.AddCodeEntry(common.BytesToHash(obj.CodeHash), path, parent, parentPath)
 		return nil
 	}
 	syncer = trie.NewSync(root, database, onAccount)

core/state/sync_test.go

@@ -134,8 +134,8 @@ func checkStateConsistency(db ethdb.Database, root common.Hash) error {
 func TestEmptyStateSync(t *testing.T) {
 	empty := common.HexToHash("56e81f171bcc55a6ff8345e692c0f86e5b48e01b996cadc001622fb5e363b421")
 	sync := NewStateSync(empty, rawdb.NewMemoryDatabase(), nil)
-	if nodes, paths, codes := sync.Missing(1); len(nodes) != 0 || len(paths) != 0 || len(codes) != 0 {
-		t.Errorf(" content requested for empty state: %v, %v, %v", nodes, paths, codes)
+	if paths, nodes, codes := sync.Missing(1); len(paths) != 0 || len(nodes) != 0 || len(codes) != 0 {
+		t.Errorf("content requested for empty state: %v, %v, %v", nodes, paths, codes)
 	}
 }
 
@@ -160,6 +160,14 @@ func TestIterativeStateSyncBatchedByPath(t *testing.T) {
 	testIterativeStateSync(t, 100, false, true)
 }
 
+// stateElement represents the element in the state trie(bytecode or trie node).
+type stateElement struct {
+	path     string
+	hash     common.Hash
+	code     common.Hash
+	syncPath trie.SyncPath
+}
+
 func testIterativeStateSync(t *testing.T, count int, commit bool, bypath bool) {
 	// Create a random state to copy
 	srcDb, srcRoot, srcAccounts := makeTestState()
@@ -172,54 +180,73 @@ func testIterativeStateSync(t *testing.T, count int, commit bool, bypath bool) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, nil)
 
-	nodes, paths, codes := sched.Missing(count)
 	var (
-		hashQueue []common.Hash
-		pathQueue []trie.SyncPath
+		nodeElements []stateElement
+		codeElements []stateElement
 	)
-	if !bypath {
-		hashQueue = append(append(hashQueue[:0], nodes...), codes...)
-	} else {
-		hashQueue = append(hashQueue[:0], codes...)
-		pathQueue = append(pathQueue[:0], paths...)
+	paths, nodes, codes := sched.Missing(count)
+	for i := 0; i < len(paths); i++ {
+		nodeElements = append(nodeElements, stateElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: trie.NewSyncPath([]byte(paths[i])),
+		})
 	}
-	for len(hashQueue)+len(pathQueue) > 0 {
-		results := make([]trie.SyncResult, len(hashQueue)+len(pathQueue))
-		for i, hash := range hashQueue {
-			data, err := srcDb.TrieDB().Node(hash)
-			if err != nil {
-				data, err = srcDb.ContractCode(common.Hash{}, hash)
-			}
+	for i := 0; i < len(codes); i++ {
+		codeElements = append(codeElements, stateElement{
+			code: codes[i],
+		})
+	}
+	for len(nodeElements)+len(codeElements) > 0 {
+		var (
+			nodeResults = make([]trie.NodeSyncResult, len(nodeElements))
+			codeResults = make([]trie.CodeSyncResult, len(codeElements))
+		)
+		for i, element := range codeElements {
+			data, err := srcDb.ContractCode(common.Hash{}, element.code)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for hash %x", hash)
+				t.Fatalf("failed to retrieve contract bytecode for hash %x", element.code)
 			}
-			results[i] = trie.SyncResult{Hash: hash, Data: data}
+			codeResults[i] = trie.CodeSyncResult{Hash: element.code, Data: data}
 		}
-		for i, path := range pathQueue {
-			if len(path) == 1 {
-				data, _, err := srcTrie.TryGetNode(path[0])
-				if err != nil {
-					t.Fatalf("failed to retrieve node data for path %x: %v", path, err)
+		for i, node := range nodeElements {
+			if bypath {
+				if len(node.syncPath) == 1 {
+					data, _, err := srcTrie.TryGetNode(node.syncPath[0])
+					if err != nil {
+						t.Fatalf("failed to retrieve node data for path %x: %v", node.syncPath[0], err)
+					}
+					nodeResults[i] = trie.NodeSyncResult{Path: node.path, Data: data}
+				} else {
+					var acc types.StateAccount
+					if err := rlp.DecodeBytes(srcTrie.Get(node.syncPath[0]), &acc); err != nil {
+						t.Fatalf("failed to decode account on path %x: %v", node.syncPath[0], err)
+					}
+					stTrie, err := trie.New(common.BytesToHash(node.syncPath[0]), acc.Root, srcDb.TrieDB())
+					if err != nil {
+						t.Fatalf("failed to retriev storage trie for path %x: %v", node.syncPath[1], err)
+					}
+					data, _, err := stTrie.TryGetNode(node.syncPath[1])
+					if err != nil {
+						t.Fatalf("failed to retrieve node data for path %x: %v", node.syncPath[1], err)
+					}
+					nodeResults[i] = trie.NodeSyncResult{Path: node.path, Data: data}
 				}
-				results[len(hashQueue)+i] = trie.SyncResult{Hash: crypto.Keccak256Hash(data), Data: data}
 			} else {
-				var acc types.StateAccount
-				if err := rlp.DecodeBytes(srcTrie.Get(path[0]), &acc); err != nil {
-					t.Fatalf("failed to decode account on path %x: %v", path, err)
-				}
-				stTrie, err := trie.New(common.BytesToHash(path[0]), acc.Root, srcDb.TrieDB())
+				data, err := srcDb.TrieDB().Node(node.hash)
 				if err != nil {
-					t.Fatalf("failed to retriev storage trie for path %x: %v", path, err)
+					t.Fatalf("failed to retrieve node data for key %v", []byte(node.path))
 				}
-				data, _, err := stTrie.TryGetNode(path[1])
-				if err != nil {
-					t.Fatalf("failed to retrieve node data for path %x: %v", path, err)
-				}
-				results[len(hashQueue)+i] = trie.SyncResult{Hash: crypto.Keccak256Hash(data), Data: data}
+				nodeResults[i] = trie.NodeSyncResult{Path: node.path, Data: data}
 			}
 		}
-		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+		for _, result := range codeResults {
+			if err := sched.ProcessCode(result); err != nil {
+				t.Errorf("failed to process result %v", err)
+			}
+		}
+		for _, result := range nodeResults {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Errorf("failed to process result %v", err)
 			}
 		}
@@ -229,12 +256,20 @@ func testIterativeStateSync(t *testing.T, count int, commit bool, bypath bool) {
 		}
 		batch.Write()
 
-		nodes, paths, codes = sched.Missing(count)
-		if !bypath {
-			hashQueue = append(append(hashQueue[:0], nodes...), codes...)
-		} else {
-			hashQueue = append(hashQueue[:0], codes...)
-			pathQueue = append(pathQueue[:0], paths...)
+		paths, nodes, codes = sched.Missing(count)
+		nodeElements = nodeElements[:0]
+		for i := 0; i < len(paths); i++ {
+			nodeElements = append(nodeElements, stateElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: trie.NewSyncPath([]byte(paths[i])),
+			})
+		}
+		codeElements = codeElements[:0]
+		for i := 0; i < len(codes); i++ {
+			codeElements = append(codeElements, stateElement{
+				code: codes[i],
+			})
 		}
 	}
 	// Cross check that the two states are in sync
@@ -251,26 +286,58 @@ func TestIterativeDelayedStateSync(t *testing.T) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, nil)
 
-	nodes, _, codes := sched.Missing(0)
-	queue := append(append([]common.Hash{}, nodes...), codes...)
-
-	for len(queue) > 0 {
+	var (
+		nodeElements []stateElement
+		codeElements []stateElement
+	)
+	paths, nodes, codes := sched.Missing(0)
+	for i := 0; i < len(paths); i++ {
+		nodeElements = append(nodeElements, stateElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: trie.NewSyncPath([]byte(paths[i])),
+		})
+	}
+	for i := 0; i < len(codes); i++ {
+		codeElements = append(codeElements, stateElement{
+			code: codes[i],
+		})
+	}
+	for len(nodeElements)+len(codeElements) > 0 {
 		// Sync only half of the scheduled nodes
-		results := make([]trie.SyncResult, len(queue)/2+1)
-		for i, hash := range queue[:len(results)] {
-			data, err := srcDb.TrieDB().Node(hash)
-			if err != nil {
-				data, err = srcDb.ContractCode(common.Hash{}, hash)
+		var nodeProcessd int
+		var codeProcessd int
+		if len(codeElements) > 0 {
+			codeResults := make([]trie.CodeSyncResult, len(codeElements)/2+1)
+			for i, element := range codeElements[:len(codeResults)] {
+				data, err := srcDb.ContractCode(common.Hash{}, element.code)
+				if err != nil {
+					t.Fatalf("failed to retrieve contract bytecode for %x", element.code)
+				}
+				codeResults[i] = trie.CodeSyncResult{Hash: element.code, Data: data}
 			}
-			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x", hash)
+			for _, result := range codeResults {
+				if err := sched.ProcessCode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
-			results[i] = trie.SyncResult{Hash: hash, Data: data}
+			codeProcessd = len(codeResults)
 		}
-		for _, result := range results {
-			if err := sched.Process(result); err != nil {
-				t.Fatalf("failed to process result %v", err)
+		if len(nodeElements) > 0 {
+			nodeResults := make([]trie.NodeSyncResult, len(nodeElements)/2+1)
+			for i, element := range nodeElements[:len(nodeResults)] {
+				data, err := srcDb.TrieDB().Node(element.hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve contract bytecode for %x", element.code)
+				}
+				nodeResults[i] = trie.NodeSyncResult{Path: element.path, Data: data}
+			}
+			for _, result := range nodeResults {
+				if err := sched.ProcessNode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
+			nodeProcessd = len(nodeResults)
 		}
 		batch := dstDb.NewBatch()
 		if err := sched.Commit(batch); err != nil {
@@ -278,8 +345,21 @@ func TestIterativeDelayedStateSync(t *testing.T) {
 		}
 		batch.Write()
 
-		nodes, _, codes = sched.Missing(0)
-		queue = append(append(queue[len(results):], nodes...), codes...)
+		paths, nodes, codes = sched.Missing(0)
+		nodeElements = nodeElements[nodeProcessd:]
+		for i := 0; i < len(paths); i++ {
+			nodeElements = append(nodeElements, stateElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: trie.NewSyncPath([]byte(paths[i])),
+			})
+		}
+		codeElements = codeElements[codeProcessd:]
+		for i := 0; i < len(codes); i++ {
+			codeElements = append(codeElements, stateElement{
+				code: codes[i],
+			})
+		}
 	}
 	// Cross check that the two states are in sync
 	checkStateAccounts(t, dstDb, srcRoot, srcAccounts)
@@ -299,40 +379,70 @@ func testIterativeRandomStateSync(t *testing.T, count int) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, nil)
 
-	queue := make(map[common.Hash]struct{})
-	nodes, _, codes := sched.Missing(count)
-	for _, hash := range append(nodes, codes...) {
-		queue[hash] = struct{}{}
+	nodeQueue := make(map[string]stateElement)
+	codeQueue := make(map[common.Hash]struct{})
+	paths, nodes, codes := sched.Missing(count)
+	for i, path := range paths {
+		nodeQueue[path] = stateElement{
+			path:     path,
+			hash:     nodes[i],
+			syncPath: trie.NewSyncPath([]byte(path)),
+		}
 	}
-	for len(queue) > 0 {
+	for _, hash := range codes {
+		codeQueue[hash] = struct{}{}
+	}
+	for len(nodeQueue)+len(codeQueue) > 0 {
 		// Fetch all the queued nodes in a random order
-		results := make([]trie.SyncResult, 0, len(queue))
-		for hash := range queue {
-			data, err := srcDb.TrieDB().Node(hash)
-			if err != nil {
-				data, err = srcDb.ContractCode(common.Hash{}, hash)
+		if len(codeQueue) > 0 {
+			results := make([]trie.CodeSyncResult, 0, len(codeQueue))
+			for hash := range codeQueue {
+				data, err := srcDb.ContractCode(common.Hash{}, hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for %x", hash)
+				}
+				results = append(results, trie.CodeSyncResult{Hash: hash, Data: data})
 			}
-			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x", hash)
+			for _, result := range results {
+				if err := sched.ProcessCode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
-			results = append(results, trie.SyncResult{Hash: hash, Data: data})
 		}
-		// Feed the retrieved results back and queue new tasks
-		for _, result := range results {
-			if err := sched.Process(result); err != nil {
-				t.Fatalf("failed to process result %v", err)
+		if len(nodeQueue) > 0 {
+			results := make([]trie.NodeSyncResult, 0, len(nodeQueue))
+			for path, element := range nodeQueue {
+				data, err := srcDb.TrieDB().Node(element.hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for %x %v %v", element.hash, []byte(element.path), element.path)
+				}
+				results = append(results, trie.NodeSyncResult{Path: path, Data: data})
+			}
+			for _, result := range results {
+				if err := sched.ProcessNode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
 		}
+		// Feed the retrieved results back and queue new tasks
 		batch := dstDb.NewBatch()
 		if err := sched.Commit(batch); err != nil {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
 
-		queue = make(map[common.Hash]struct{})
-		nodes, _, codes = sched.Missing(count)
-		for _, hash := range append(nodes, codes...) {
-			queue[hash] = struct{}{}
+		nodeQueue = make(map[string]stateElement)
+		codeQueue = make(map[common.Hash]struct{})
+		paths, nodes, codes := sched.Missing(count)
+		for i, path := range paths {
+			nodeQueue[path] = stateElement{
+				path:     path,
+				hash:     nodes[i],
+				syncPath: trie.NewSyncPath([]byte(path)),
+			}
+		}
+		for _, hash := range codes {
+			codeQueue[hash] = struct{}{}
 		}
 	}
 	// Cross check that the two states are in sync
@@ -349,34 +459,62 @@ func TestIterativeRandomDelayedStateSync(t *testing.T) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, nil)
 
-	queue := make(map[common.Hash]struct{})
-	nodes, _, codes := sched.Missing(0)
-	for _, hash := range append(nodes, codes...) {
-		queue[hash] = struct{}{}
+	nodeQueue := make(map[string]stateElement)
+	codeQueue := make(map[common.Hash]struct{})
+	paths, nodes, codes := sched.Missing(0)
+	for i, path := range paths {
+		nodeQueue[path] = stateElement{
+			path:     path,
+			hash:     nodes[i],
+			syncPath: trie.NewSyncPath([]byte(path)),
+		}
+	}
+	for _, hash := range codes {
+		codeQueue[hash] = struct{}{}
 	}
-	for len(queue) > 0 {
+	for len(nodeQueue)+len(codeQueue) > 0 {
 		// Sync only half of the scheduled nodes, even those in random order
-		results := make([]trie.SyncResult, 0, len(queue)/2+1)
-		for hash := range queue {
-			delete(queue, hash)
+		if len(codeQueue) > 0 {
+			results := make([]trie.CodeSyncResult, 0, len(codeQueue)/2+1)
+			for hash := range codeQueue {
+				delete(codeQueue, hash)
 
-			data, err := srcDb.TrieDB().Node(hash)
-			if err != nil {
-				data, err = srcDb.ContractCode(common.Hash{}, hash)
+				data, err := srcDb.ContractCode(common.Hash{}, hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for %x", hash)
+				}
+				results = append(results, trie.CodeSyncResult{Hash: hash, Data: data})
+
+				if len(results) >= cap(results) {
+					break
+				}
 			}
-			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x", hash)
+			for _, result := range results {
+				if err := sched.ProcessCode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
-			results = append(results, trie.SyncResult{Hash: hash, Data: data})
+		}
+		if len(nodeQueue) > 0 {
+			results := make([]trie.NodeSyncResult, 0, len(nodeQueue)/2+1)
+			for path, element := range nodeQueue {
+				delete(nodeQueue, path)
 
-			if len(results) >= cap(results) {
-				break
+				data, err := srcDb.TrieDB().Node(element.hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for %x", element.hash)
+				}
+				results = append(results, trie.NodeSyncResult{Path: path, Data: data})
+
+				if len(results) >= cap(results) {
+					break
+				}
 			}
-		}
-		// Feed the retrieved results back and queue new tasks
-		for _, result := range results {
-			if err := sched.Process(result); err != nil {
-				t.Fatalf("failed to process result %v", err)
+			// Feed the retrieved results back and queue new tasks
+			for _, result := range results {
+				if err := sched.ProcessNode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
 		}
 		batch := dstDb.NewBatch()
@@ -384,12 +522,17 @@ func TestIterativeRandomDelayedStateSync(t *testing.T) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
-		for _, result := range results {
-			delete(queue, result.Hash)
+
+		paths, nodes, codes := sched.Missing(0)
+		for i, path := range paths {
+			nodeQueue[path] = stateElement{
+				path:     path,
+				hash:     nodes[i],
+				syncPath: trie.NewSyncPath([]byte(path)),
+			}
 		}
-		nodes, _, codes = sched.Missing(0)
-		for _, hash := range append(nodes, codes...) {
-			queue[hash] = struct{}{}
+		for _, hash := range codes {
+			codeQueue[hash] = struct{}{}
 		}
 	}
 	// Cross check that the two states are in sync
@@ -416,28 +559,62 @@ func TestIncompleteStateSync(t *testing.T) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, nil)
 
-	var added []common.Hash
-
-	nodes, _, codes := sched.Missing(1)
-	queue := append(append([]common.Hash{}, nodes...), codes...)
-
-	for len(queue) > 0 {
+	var (
+		addedCodes []common.Hash
+		addedNodes []common.Hash
+	)
+	nodeQueue := make(map[string]stateElement)
+	codeQueue := make(map[common.Hash]struct{})
+	paths, nodes, codes := sched.Missing(1)
+	for i, path := range paths {
+		nodeQueue[path] = stateElement{
+			path:     path,
+			hash:     nodes[i],
+			syncPath: trie.NewSyncPath([]byte(path)),
+		}
+	}
+	for _, hash := range codes {
+		codeQueue[hash] = struct{}{}
+	}
+	for len(nodeQueue)+len(codeQueue) > 0 {
 		// Fetch a batch of state nodes
-		results := make([]trie.SyncResult, len(queue))
-		for i, hash := range queue {
-			data, err := srcDb.TrieDB().Node(hash)
-			if err != nil {
-				data, err = srcDb.ContractCode(common.Hash{}, hash)
+		if len(codeQueue) > 0 {
+			results := make([]trie.CodeSyncResult, 0, len(codeQueue))
+			for hash := range codeQueue {
+				data, err := srcDb.ContractCode(common.Hash{}, hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for %x", hash)
+				}
+				results = append(results, trie.CodeSyncResult{Hash: hash, Data: data})
+				addedCodes = append(addedCodes, hash)
 			}
-			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x", hash)
+			// Process each of the state nodes
+			for _, result := range results {
+				if err := sched.ProcessCode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
-			results[i] = trie.SyncResult{Hash: hash, Data: data}
 		}
-		// Process each of the state nodes
-		for _, result := range results {
-			if err := sched.Process(result); err != nil {
-				t.Fatalf("failed to process result %v", err)
+		var nodehashes []common.Hash
+		if len(nodeQueue) > 0 {
+			results := make([]trie.NodeSyncResult, 0, len(nodeQueue))
+			for key, element := range nodeQueue {
+				data, err := srcDb.TrieDB().Node(element.hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for %x", element.hash)
+				}
+				results = append(results, trie.NodeSyncResult{Path: key, Data: data})
+
+				if element.hash != srcRoot {
+					addedNodes = append(addedNodes, element.hash)
+				}
+				nodehashes = append(nodehashes, element.hash)
+			}
+			// Process each of the state nodes
+			for _, result := range results {
+				if err := sched.ProcessNode(result); err != nil {
+					t.Fatalf("failed to process result %v", err)
+				}
 			}
 		}
 		batch := dstDb.NewBatch()
@@ -445,43 +622,44 @@ func TestIncompleteStateSync(t *testing.T) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
-		for _, result := range results {
-			added = append(added, result.Hash)
-			// Check that all known sub-tries added so far are complete or missing entirely.
-			if _, ok := isCode[result.Hash]; ok {
-				continue
-			}
+
+		for _, root := range nodehashes {
 			// Can't use checkStateConsistency here because subtrie keys may have odd
 			// length and crash in LeafKey.
-			if err := checkTrieConsistency(dstDb, result.Hash); err != nil {
+			if err := checkTrieConsistency(dstDb, root); err != nil {
 				t.Fatalf("state inconsistent: %v", err)
 			}
 		}
 		// Fetch the next batch to retrieve
-		nodes, _, codes = sched.Missing(1)
-		queue = append(append(queue[:0], nodes...), codes...)
+		nodeQueue = make(map[string]stateElement)
+		codeQueue = make(map[common.Hash]struct{})
+		paths, nodes, codes := sched.Missing(1)
+		for i, path := range paths {
+			nodeQueue[path] = stateElement{
+				path:     path,
+				hash:     nodes[i],
+				syncPath: trie.NewSyncPath([]byte(path)),
+			}
+		}
+		for _, hash := range codes {
+			codeQueue[hash] = struct{}{}
+		}
 	}
 	// Sanity check that removing any node from the database is detected
-	for _, node := range added[1:] {
-		var (
-			key     = node.Bytes()
-			_, code = isCode[node]
-			val     []byte
-		)
-		if code {
-			val = rawdb.ReadCode(dstDb, node)
-			rawdb.DeleteCode(dstDb, node)
-		} else {
-			val = rawdb.ReadTrieNode(dstDb, node)
-			rawdb.DeleteTrieNode(dstDb, node)
+	for _, node := range addedCodes {
+		val := rawdb.ReadCode(dstDb, node)
+		rawdb.DeleteCode(dstDb, node)
+		if err := checkStateConsistency(dstDb, srcRoot); err == nil {
+			t.Errorf("trie inconsistency not caught, missing: %x", node)
 		}
-		if err := checkStateConsistency(dstDb, added[0]); err == nil {
-			t.Fatalf("trie inconsistency not caught, missing: %x", key)
-		}
-		if code {
-			rawdb.WriteCode(dstDb, node, val)
-		} else {
-			rawdb.WriteTrieNode(dstDb, node, val)
+		rawdb.WriteCode(dstDb, node, val)
+	}
+	for _, node := range addedNodes {
+		val := rawdb.ReadTrieNode(dstDb, node)
+		rawdb.DeleteTrieNode(dstDb, node)
+		if err := checkStateConsistency(dstDb, srcRoot); err == nil {
+			t.Errorf("trie inconsistency not caught, missing: %v", node.Hex())
 		}
+		rawdb.WriteTrieNode(dstDb, node, val)
 	}
 }

eth/downloader/skeleton_test.go

@@ -21,7 +21,6 @@ import (
 	"errors"
 	"fmt"
 	"math/big"
-	"os"
 	"sync/atomic"
 	"testing"
 	"time"
@@ -515,7 +514,7 @@ func TestSkeletonSyncExtend(t *testing.T) {
 // Tests that the skeleton sync correctly retrieves headers from one or more
 // peers without duplicates or other strange side effects.
 func TestSkeletonSyncRetrievals(t *testing.T) {
-	log.Root().SetHandler(log.LvlFilterHandler(log.LvlTrace, log.StreamHandler(os.Stderr, log.TerminalFormat(true))))
+	//log.Root().SetHandler(log.LvlFilterHandler(log.LvlTrace, log.StreamHandler(os.Stderr, log.TerminalFormat(true))))
 
 	// Since skeleton headers don't need to be meaningful, beyond a parent hash
 	// progression, create a long fake chain to test with.

eth/protocols/snap/sort_test.go

@@ -22,7 +22,6 @@ import (
 	"testing"
 
 	"github.com/ethereum/go-ethereum/common"
-	"github.com/ethereum/go-ethereum/trie"
 )
 
 func hexToNibbles(s string) []byte {
@@ -38,22 +37,17 @@ func hexToNibbles(s string) []byte {
 }
 
 func TestRequestSorting(t *testing.T) {
-
 	//   - Path 0x9  -> {0x19}
 	//   - Path 0x99 -> {0x0099}
 	//   - Path 0x01234567890123456789012345678901012345678901234567890123456789019  -> {0x0123456789012345678901234567890101234567890123456789012345678901, 0x19}
 	//   - Path 0x012345678901234567890123456789010123456789012345678901234567890199 -> {0x0123456789012345678901234567890101234567890123456789012345678901, 0x0099}
-	var f = func(path string) (trie.SyncPath, TrieNodePathSet, common.Hash) {
+	var f = func(path string) string {
 		data := hexToNibbles(path)
-		sp := trie.NewSyncPath(data)
-		tnps := TrieNodePathSet([][]byte(sp))
-		hash := common.Hash{}
-		return sp, tnps, hash
+		return string(data)
 	}
 	var (
-		hashes   []common.Hash
-		paths    []trie.SyncPath
-		pathsets []TrieNodePathSet
+		hashes []common.Hash
+		paths  []string
 	)
 	for _, x := range []string{
 		"0x9",
@@ -67,15 +61,14 @@ func TestRequestSorting(t *testing.T) {
 		"0x01234567890123456789012345678901012345678901234567890123456789010",
 		"0x01234567890123456789012345678901012345678901234567890123456789011",
 	} {
-		sp, _, hash := f(x)
-		hashes = append(hashes, hash)
-		paths = append(paths, sp)
+		paths = append(paths, f(x))
+		hashes = append(hashes, common.Hash{})
 	}
-	_, paths, pathsets = sortByAccountPath(hashes, paths)
+	_, _, syncPaths, pathsets := sortByAccountPath(paths, hashes)
 	{
 		var b = new(bytes.Buffer)
-		for i := 0; i < len(paths); i++ {
-			fmt.Fprintf(b, "\n%d. paths %x", i, paths[i])
+		for i := 0; i < len(syncPaths); i++ {
+			fmt.Fprintf(b, "\n%d. paths %x", i, syncPaths[i])
 		}
 		want := `
 0. paths [0099]

eth/protocols/snap/sync.go

@@ -230,8 +230,8 @@ type trienodeHealRequest struct {
 	timeout *time.Timer                // Timer to track delivery timeout
 	stale   chan struct{}              // Channel to signal the request was dropped
 
-	hashes []common.Hash   // Trie node hashes to validate responses
-	paths  []trie.SyncPath // Trie node paths requested for rescheduling
+	paths  []string      // Trie node paths for identifying trie node
+	hashes []common.Hash // Trie node hashes to validate responses
 
 	task *healTask // Task which this request is filling (only access fields through the runloop!!)
 }
@@ -240,9 +240,9 @@ type trienodeHealRequest struct {
 type trienodeHealResponse struct {
 	task *healTask // Task which this request is filling
 
-	hashes []common.Hash   // Hashes of the trie nodes to avoid double hashing
-	paths  []trie.SyncPath // Trie node paths requested for rescheduling missing ones
-	nodes  [][]byte        // Actual trie nodes to store into the database (nil = missing)
+	paths  []string      // Paths of the trie nodes
+	hashes []common.Hash // Hashes of the trie nodes to avoid double hashing
+	nodes  [][]byte      // Actual trie nodes to store into the database (nil = missing)
 }
 
 // bytecodeHealRequest tracks a pending bytecode request to ensure responses are to
@@ -321,8 +321,8 @@ type storageTask struct {
 type healTask struct {
 	scheduler *trie.Sync // State trie sync scheduler defining the tasks
 
-	trieTasks map[common.Hash]trie.SyncPath // Set of trie node tasks currently queued for retrieval
-	codeTasks map[common.Hash]struct{}      // Set of byte code tasks currently queued for retrieval
+	trieTasks map[string]common.Hash   // Set of trie node tasks currently queued for retrieval, indexed by node path
+	codeTasks map[common.Hash]struct{} // Set of byte code tasks currently queued for retrieval, indexed by code hash
 }
 
 // SyncProgress is a database entry to allow suspending and resuming a snapshot state
@@ -540,7 +540,7 @@ func (s *Syncer) Unregister(id string) error {
 	return nil
 }
 
-// Sync starts (or resumes a previous) sync cycle to iterate over an state trie
+// Sync starts (or resumes a previous) sync cycle to iterate over a state trie
 // with the given root and reconstruct the nodes based on the snapshot leaves.
 // Previously downloaded segments will not be redownloaded of fixed, rather any
 // errors will be healed after the leaves are fully accumulated.
@@ -551,7 +551,7 @@ func (s *Syncer) Sync(root common.Hash, cancel chan struct{}) error {
 	s.root = root
 	s.healer = &healTask{
 		scheduler: state.NewStateSync(root, s.db, s.onHealState),
-		trieTasks: make(map[common.Hash]trie.SyncPath),
+		trieTasks: make(map[string]common.Hash),
 		codeTasks: make(map[common.Hash]struct{}),
 	}
 	s.statelessPeers = make(map[string]struct{})
@@ -743,7 +743,7 @@ func (s *Syncer) loadSyncStatus() {
 			return
 		}
 	}
-	// Either we've failed to decode the previus state, or there was none.
+	// Either we've failed to decode the previous state, or there was none.
 	// Start a fresh sync by chunking up the account range and scheduling
 	// them for retrieval.
 	s.tasks = nil
@@ -1280,9 +1280,9 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 			want = maxTrieRequestCount + maxCodeRequestCount
 		)
 		if have < want {
-			nodes, paths, codes := s.healer.scheduler.Missing(want - have)
-			for i, hash := range nodes {
-				s.healer.trieTasks[hash] = paths[i]
+			paths, hashes, codes := s.healer.scheduler.Missing(want - have)
+			for i, path := range paths {
+				s.healer.trieTasks[path] = hashes[i]
 			}
 			for _, hash := range codes {
 				s.healer.codeTasks[hash] = struct{}{}
@@ -1323,21 +1323,20 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 		}
 		var (
 			hashes   = make([]common.Hash, 0, cap)
-			paths    = make([]trie.SyncPath, 0, cap)
+			paths    = make([]string, 0, cap)
 			pathsets = make([]TrieNodePathSet, 0, cap)
 		)
-		for hash, pathset := range s.healer.trieTasks {
-			delete(s.healer.trieTasks, hash)
+		for path, hash := range s.healer.trieTasks {
+			delete(s.healer.trieTasks, path)
 
+			paths = append(paths, path)
 			hashes = append(hashes, hash)
-			paths = append(paths, pathset)
-
-			if len(hashes) >= cap {
+			if len(paths) >= cap {
 				break
 			}
 		}
 		// Group requests by account hash
-		hashes, paths, pathsets = sortByAccountPath(hashes, paths)
+		paths, hashes, _, pathsets = sortByAccountPath(paths, hashes)
 		req := &trienodeHealRequest{
 			peer:    idle,
 			id:      reqid,
@@ -1346,8 +1345,8 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 			revert:  fail,
 			cancel:  cancel,
 			stale:   make(chan struct{}),
-			hashes:  hashes,
 			paths:   paths,
+			hashes:  hashes,
 			task:    s.healer,
 		}
 		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
@@ -1405,9 +1404,9 @@ func (s *Syncer) assignBytecodeHealTasks(success chan *bytecodeHealResponse, fai
 			want = maxTrieRequestCount + maxCodeRequestCount
 		)
 		if have < want {
-			nodes, paths, codes := s.healer.scheduler.Missing(want - have)
-			for i, hash := range nodes {
-				s.healer.trieTasks[hash] = paths[i]
+			paths, hashes, codes := s.healer.scheduler.Missing(want - have)
+			for i, path := range paths {
+				s.healer.trieTasks[path] = hashes[i]
 			}
 			for _, hash := range codes {
 				s.healer.codeTasks[hash] = struct{}{}
@@ -1703,10 +1702,10 @@ func (s *Syncer) revertTrienodeHealRequest(req *trienodeHealRequest) {
 	s.lock.Unlock()
 
 	// If there's a timeout timer still running, abort it and mark the trie node
-	// retrievals as not-pending, ready for resheduling
+	// retrievals as not-pending, ready for rescheduling
 	req.timeout.Stop()
-	for i, hash := range req.hashes {
-		req.task.trieTasks[hash] = req.paths[i]
+	for i, path := range req.paths {
+		req.task.trieTasks[path] = req.hashes[i]
 	}
 }
 
@@ -2096,14 +2095,14 @@ func (s *Syncer) processTrienodeHealResponse(res *trienodeHealResponse) {
 
 		// If the trie node was not delivered, reschedule it
 		if node == nil {
-			res.task.trieTasks[hash] = res.paths[i]
+			res.task.trieTasks[res.paths[i]] = res.hashes[i]
 			continue
 		}
 		// Push the trie node into the state syncer
 		s.trienodeHealSynced++
 		s.trienodeHealBytes += common.StorageSize(len(node))
 
-		err := s.healer.scheduler.Process(trie.SyncResult{Hash: hash, Data: node})
+		err := s.healer.scheduler.ProcessNode(trie.NodeSyncResult{Path: res.paths[i], Data: node})
 		switch err {
 		case nil:
 		case trie.ErrAlreadyProcessed:
@@ -2139,7 +2138,7 @@ func (s *Syncer) processBytecodeHealResponse(res *bytecodeHealResponse) {
 		s.bytecodeHealSynced++
 		s.bytecodeHealBytes += common.StorageSize(len(node))
 
-		err := s.healer.scheduler.Process(trie.SyncResult{Hash: hash, Data: node})
+		err := s.healer.scheduler.ProcessCode(trie.CodeSyncResult{Hash: hash, Data: node})
 		switch err {
 		case nil:
 		case trie.ErrAlreadyProcessed:
@@ -2666,9 +2665,9 @@ func (s *Syncer) OnTrieNodes(peer SyncPeer, id uint64, trienodes [][]byte) error
 	}
 	// Response validated, send it to the scheduler for filling
 	response := &trienodeHealResponse{
+		paths:  req.paths,
 		task:   req.task,
 		hashes: req.hashes,
-		paths:  req.paths,
 		nodes:  nodes,
 	}
 	select {
@@ -2913,8 +2912,9 @@ func (s *capacitySort) Swap(i, j int) {
 // healRequestSort implements the Sort interface, allowing sorting trienode
 // heal requests, which is a prerequisite for merging storage-requests.
 type healRequestSort struct {
-	hashes []common.Hash
-	paths  []trie.SyncPath
+	paths     []string
+	hashes    []common.Hash
+	syncPaths []trie.SyncPath
 }
 
 func (t *healRequestSort) Len() int {
@@ -2922,8 +2922,8 @@ func (t *healRequestSort) Len() int {
 }
 
 func (t *healRequestSort) Less(i, j int) bool {
-	a := t.paths[i]
-	b := t.paths[j]
+	a := t.syncPaths[i]
+	b := t.syncPaths[j]
 	switch bytes.Compare(a[0], b[0]) {
 	case -1:
 		return true
@@ -2944,8 +2944,9 @@ func (t *healRequestSort) Less(i, j int) bool {
 }
 
 func (t *healRequestSort) Swap(i, j int) {
-	t.hashes[i], t.hashes[j] = t.hashes[j], t.hashes[i]
 	t.paths[i], t.paths[j] = t.paths[j], t.paths[i]
+	t.hashes[i], t.hashes[j] = t.hashes[j], t.hashes[i]
+	t.syncPaths[i], t.syncPaths[j] = t.syncPaths[j], t.syncPaths[i]
 }
 
 // Merge merges the pathsets, so that several storage requests concerning the
@@ -2953,7 +2954,7 @@ func (t *healRequestSort) Swap(i, j int) {
 // OBS: This operation is moot if t has not first been sorted.
 func (t *healRequestSort) Merge() []TrieNodePathSet {
 	var result []TrieNodePathSet
-	for _, path := range t.paths {
+	for _, path := range t.syncPaths {
 		pathset := TrieNodePathSet([][]byte(path))
 		if len(path) == 1 {
 			// It's an account reference.
@@ -2962,7 +2963,7 @@ func (t *healRequestSort) Merge() []TrieNodePathSet {
 			// It's a storage reference.
 			end := len(result) - 1
 			if len(result) == 0 || !bytes.Equal(pathset[0], result[end][0]) {
-				// The account doesn't doesn't match last, create a new entry.
+				// The account doesn't match last, create a new entry.
 				result = append(result, pathset)
 			} else {
 				// It's the same account as the previous one, add to the storage
@@ -2976,9 +2977,13 @@ func (t *healRequestSort) Merge() []TrieNodePathSet {
 
 // sortByAccountPath takes hashes and paths, and sorts them. After that, it generates
 // the TrieNodePaths and merges paths which belongs to the same account path.
-func sortByAccountPath(hashes []common.Hash, paths []trie.SyncPath) ([]common.Hash, []trie.SyncPath, []TrieNodePathSet) {
-	n := &healRequestSort{hashes, paths}
+func sortByAccountPath(paths []string, hashes []common.Hash) ([]string, []common.Hash, []trie.SyncPath, []TrieNodePathSet) {
+	var syncPaths []trie.SyncPath
+	for _, path := range paths {
+		syncPaths = append(syncPaths, trie.NewSyncPath([]byte(path)))
+	}
+	n := &healRequestSort{paths, hashes, syncPaths}
 	sort.Sort(n)
 	pathsets := n.Merge()
-	return n.hashes, n.paths, pathsets
+	return n.paths, n.hashes, n.syncPaths, pathsets
 }

les/downloader/statesync.go

@@ -34,7 +34,7 @@ import (
 // a single data retrieval network packet.
 type stateReq struct {
 	nItems    uint16                    // Number of items requested for download (max is 384, so uint16 is sufficient)
-	trieTasks map[common.Hash]*trieTask // Trie node download tasks to track previous attempts
+	trieTasks map[string]*trieTask      // Trie node download tasks to track previous attempts
 	codeTasks map[common.Hash]*codeTask // Byte code download tasks to track previous attempts
 	timeout   time.Duration             // Maximum round trip time for this to complete
 	timer     *time.Timer               // Timer to fire when the RTT timeout expires
@@ -263,8 +263,8 @@ type stateSync struct {
 	sched  *trie.Sync         // State trie sync scheduler defining the tasks
 	keccak crypto.KeccakState // Keccak256 hasher to verify deliveries with
 
-	trieTasks map[common.Hash]*trieTask // Set of trie node tasks currently queued for retrieval
-	codeTasks map[common.Hash]*codeTask // Set of byte code tasks currently queued for retrieval
+	trieTasks map[string]*trieTask      // Set of trie node tasks currently queued for retrieval, indexed by path
+	codeTasks map[common.Hash]*codeTask // Set of byte code tasks currently queued for retrieval, indexed by hash
 
 	numUncommitted   int
 	bytesUncommitted int
@@ -281,6 +281,7 @@ type stateSync struct {
 // trieTask represents a single trie node download task, containing a set of
 // peers already attempted retrieval from to detect stalled syncs and abort.
 type trieTask struct {
+	hash     common.Hash
 	path     [][]byte
 	attempts map[string]struct{}
 }
@@ -299,7 +300,7 @@ func newStateSync(d *Downloader, root common.Hash) *stateSync {
 		root:      root,
 		sched:     state.NewStateSync(root, d.stateDB, nil),
 		keccak:    sha3.NewLegacyKeccak256().(crypto.KeccakState),
-		trieTasks: make(map[common.Hash]*trieTask),
+		trieTasks: make(map[string]*trieTask),
 		codeTasks: make(map[common.Hash]*codeTask),
 		deliver:   make(chan *stateReq),
 		cancel:    make(chan struct{}),
@@ -455,10 +456,11 @@ func (s *stateSync) assignTasks() {
 func (s *stateSync) fillTasks(n int, req *stateReq) (nodes []common.Hash, paths []trie.SyncPath, codes []common.Hash) {
 	// Refill available tasks from the scheduler.
 	if fill := n - (len(s.trieTasks) + len(s.codeTasks)); fill > 0 {
-		nodes, paths, codes := s.sched.Missing(fill)
-		for i, hash := range nodes {
-			s.trieTasks[hash] = &trieTask{
-				path:     paths[i],
+		paths, hashes, codes := s.sched.Missing(fill)
+		for i, path := range paths {
+			s.trieTasks[path] = &trieTask{
+				hash:     hashes[i],
+				path:     trie.NewSyncPath([]byte(path)),
 				attempts: make(map[string]struct{}),
 			}
 		}
@@ -474,7 +476,7 @@ func (s *stateSync) fillTasks(n int, req *stateReq) (nodes []common.Hash, paths
 	paths = make([]trie.SyncPath, 0, n)
 	codes = make([]common.Hash, 0, n)
 
-	req.trieTasks = make(map[common.Hash]*trieTask, n)
+	req.trieTasks = make(map[string]*trieTask, n)
 	req.codeTasks = make(map[common.Hash]*codeTask, n)
 
 	for hash, t := range s.codeTasks {
@@ -492,7 +494,7 @@ func (s *stateSync) fillTasks(n int, req *stateReq) (nodes []common.Hash, paths
 		req.codeTasks[hash] = t
 		delete(s.codeTasks, hash)
 	}
-	for hash, t := range s.trieTasks {
+	for path, t := range s.trieTasks {
 		// Stop when we've gathered enough requests
 		if len(nodes)+len(codes) == n {
 			break
@@ -504,11 +506,11 @@ func (s *stateSync) fillTasks(n int, req *stateReq) (nodes []common.Hash, paths
 		// Assign the request to this peer
 		t.attempts[req.peer.id] = struct{}{}
 
-		nodes = append(nodes, hash)
+		nodes = append(nodes, t.hash)
 		paths = append(paths, t.path)
 
-		req.trieTasks[hash] = t
-		delete(s.trieTasks, hash)
+		req.trieTasks[path] = t
+		delete(s.trieTasks, path)
 	}
 	req.nItems = uint16(len(nodes) + len(codes))
 	return nodes, paths, codes
@@ -530,7 +532,7 @@ func (s *stateSync) process(req *stateReq) (int, error) {
 
 	// Iterate over all the delivered data and inject one-by-one into the trie
 	for _, blob := range req.response {
-		hash, err := s.processNodeData(blob)
+		hash, err := s.processNodeData(req.trieTasks, req.codeTasks, blob)
 		switch err {
 		case nil:
 			s.numUncommitted++
@@ -543,13 +545,10 @@ func (s *stateSync) process(req *stateReq) (int, error) {
 		default:
 			return successful, fmt.Errorf("invalid state node %s: %v", hash.TerminalString(), err)
 		}
-		// Delete from both queues (one delivery is enough for the syncer)
-		delete(req.trieTasks, hash)
-		delete(req.codeTasks, hash)
 	}
 	// Put unfulfilled tasks back into the retry queue
 	npeers := s.d.peers.Len()
-	for hash, task := range req.trieTasks {
+	for path, task := range req.trieTasks {
 		// If the node did deliver something, missing items may be due to a protocol
 		// limit or a previous timeout + delayed delivery. Both cases should permit
 		// the node to retry the missing items (to avoid single-peer stalls).
@@ -559,10 +558,10 @@ func (s *stateSync) process(req *stateReq) (int, error) {
 		// If we've requested the node too many times already, it may be a malicious
 		// sync where nobody has the right data. Abort.
 		if len(task.attempts) >= npeers {
-			return successful, fmt.Errorf("trie node %s failed with all peers (%d tries, %d peers)", hash.TerminalString(), len(task.attempts), npeers)
+			return successful, fmt.Errorf("trie node %s failed with all peers (%d tries, %d peers)", task.hash.TerminalString(), len(task.attempts), npeers)
 		}
 		// Missing item, place into the retry queue.
-		s.trieTasks[hash] = task
+		s.trieTasks[path] = task
 	}
 	for hash, task := range req.codeTasks {
 		// If the node did deliver something, missing items may be due to a protocol
@@ -585,13 +584,35 @@ func (s *stateSync) process(req *stateReq) (int, error) {
 // processNodeData tries to inject a trie node data blob delivered from a remote
 // peer into the state trie, returning whether anything useful was written or any
 // error occurred.
-func (s *stateSync) processNodeData(blob []byte) (common.Hash, error) {
-	res := trie.SyncResult{Data: blob}
+//
+// If multiple requests correspond to the same hash, this method will inject the
+// blob as a result for the first one only, leaving the remaining duplicates to
+// be fetched again.
+func (s *stateSync) processNodeData(nodeTasks map[string]*trieTask, codeTasks map[common.Hash]*codeTask, blob []byte) (common.Hash, error) {
+	var hash common.Hash
 	s.keccak.Reset()
 	s.keccak.Write(blob)
-	s.keccak.Read(res.Hash[:])
-	err := s.sched.Process(res)
-	return res.Hash, err
+	s.keccak.Read(hash[:])
+
+	if _, present := codeTasks[hash]; present {
+		err := s.sched.ProcessCode(trie.CodeSyncResult{
+			Hash: hash,
+			Data: blob,
+		})
+		delete(codeTasks, hash)
+		return hash, err
+	}
+	for path, task := range nodeTasks {
+		if task.hash == hash {
+			err := s.sched.ProcessNode(trie.NodeSyncResult{
+				Path: path,
+				Data: blob,
+			})
+			delete(nodeTasks, path)
+			return hash, err
+		}
+	}
+	return common.Hash{}, trie.ErrNotRequested
 }
 
 // updateStats bumps the various state sync progress counters and displays a log

trie/committer.go

@@ -33,6 +33,7 @@ type leaf struct {
 	size int         // size of the rlp data (estimate)
 	hash common.Hash // hash of rlp data
 	node node        // the node to commit
+	path []byte      // the path from the root node
 }
 
 // committer is a type used for the trie Commit operation. A committer has some
@@ -69,7 +70,7 @@ func (c *committer) Commit(n node, db *Database) (hashNode, int, error) {
 	if db == nil {
 		return nil, 0, errors.New("no db provided")
 	}
-	h, committed, err := c.commit(n, db)
+	h, committed, err := c.commit(nil, n, db)
 	if err != nil {
 		return nil, 0, err
 	}
@@ -77,7 +78,7 @@ func (c *committer) Commit(n node, db *Database) (hashNode, int, error) {
 }
 
 // commit collapses a node down into a hash node and inserts it into the database
-func (c *committer) commit(n node, db *Database) (node, int, error) {
+func (c *committer) commit(path []byte, n node, db *Database) (node, int, error) {
 	// if this path is clean, use available cached data
 	hash, dirty := n.cache()
 	if hash != nil && !dirty {
@@ -93,7 +94,7 @@ func (c *committer) commit(n node, db *Database) (node, int, error) {
 		// otherwise it can only be hashNode or valueNode.
 		var childCommitted int
 		if _, ok := cn.Val.(*fullNode); ok {
-			childV, committed, err := c.commit(cn.Val, db)
+			childV, committed, err := c.commit(append(path, cn.Key...), cn.Val, db)
 			if err != nil {
 				return nil, 0, err
 			}
@@ -101,20 +102,20 @@ func (c *committer) commit(n node, db *Database) (node, int, error) {
 		}
 		// The key needs to be copied, since we're delivering it to database
 		collapsed.Key = hexToCompact(cn.Key)
-		hashedNode := c.store(collapsed, db)
+		hashedNode := c.store(path, collapsed, db)
 		if hn, ok := hashedNode.(hashNode); ok {
 			return hn, childCommitted + 1, nil
 		}
 		return collapsed, childCommitted, nil
 	case *fullNode:
-		hashedKids, childCommitted, err := c.commitChildren(cn, db)
+		hashedKids, childCommitted, err := c.commitChildren(path, cn, db)
 		if err != nil {
 			return nil, 0, err
 		}
 		collapsed := cn.copy()
 		collapsed.Children = hashedKids
 
-		hashedNode := c.store(collapsed, db)
+		hashedNode := c.store(path, collapsed, db)
 		if hn, ok := hashedNode.(hashNode); ok {
 			return hn, childCommitted + 1, nil
 		}
@@ -128,7 +129,7 @@ func (c *committer) commit(n node, db *Database) (node, int, error) {
 }
 
 // commitChildren commits the children of the given fullnode
-func (c *committer) commitChildren(n *fullNode, db *Database) ([17]node, int, error) {
+func (c *committer) commitChildren(path []byte, n *fullNode, db *Database) ([17]node, int, error) {
 	var (
 		committed int
 		children  [17]node
@@ -148,7 +149,7 @@ func (c *committer) commitChildren(n *fullNode, db *Database) ([17]node, int, er
 		// Commit the child recursively and store the "hashed" value.
 		// Note the returned node can be some embedded nodes, so it's
 		// possible the type is not hashNode.
-		hashed, childCommitted, err := c.commit(child, db)
+		hashed, childCommitted, err := c.commit(append(path, byte(i)), child, db)
 		if err != nil {
 			return children, 0, err
 		}
@@ -165,7 +166,7 @@ func (c *committer) commitChildren(n *fullNode, db *Database) ([17]node, int, er
 // store hashes the node n and if we have a storage layer specified, it writes
 // the key/value pair to it and tracks any node->child references as well as any
 // node->external trie references.
-func (c *committer) store(n node, db *Database) node {
+func (c *committer) store(path []byte, n node, db *Database) node {
 	// Larger nodes are replaced by their hash and stored in the database.
 	var (
 		hash, _ = n.cache()
@@ -189,6 +190,7 @@ func (c *committer) store(n node, db *Database) node {
 			size: size,
 			hash: common.BytesToHash(hash),
 			node: n,
+			path: path,
 		}
 	} else if db != nil {
 		// No leaf-callback used, but there's still a database. Do serial
@@ -213,13 +215,13 @@ func (c *committer) commitLoop(db *Database) {
 			switch n := n.(type) {
 			case *shortNode:
 				if child, ok := n.Val.(valueNode); ok {
-					c.onleaf(nil, nil, child, hash)
+					c.onleaf(nil, nil, child, hash, nil)
 				}
 			case *fullNode:
 				// For children in range [0, 15], it's impossible
 				// to contain valueNode. Only check the 17th child.
 				if n.Children[16] != nil {
-					c.onleaf(nil, nil, n.Children[16].(valueNode), hash)
+					c.onleaf(nil, nil, n.Children[16].(valueNode), hash, nil)
 				}
 			}
 		}

trie/sync.go

@@ -24,6 +24,7 @@ import (
 	"github.com/ethereum/go-ethereum/common/prque"
 	"github.com/ethereum/go-ethereum/core/rawdb"
 	"github.com/ethereum/go-ethereum/ethdb"
+	"github.com/ethereum/go-ethereum/log"
 )
 
 // ErrNotRequested is returned by the trie sync when it's requested to process a
@@ -39,19 +40,6 @@ var ErrAlreadyProcessed = errors.New("already processed")
 // memory if the node was configured with a significant number of peers.
 const maxFetchesPerDepth = 16384
 
-// request represents a scheduled or already in-flight state retrieval request.
-type request struct {
-	path []byte      // Merkle path leading to this node for prioritization
-	hash common.Hash // Hash of the node data content to retrieve
-	data []byte      // Data content of the node, cached until all subtrees complete
-	code bool        // Whether this is a code entry
-
-	parents []*request // Parent state nodes referencing this entry (notify all upon completion)
-	deps    int        // Number of dependencies before allowed to commit this node
-
-	callback LeafCallback // Callback to invoke if a leaf node it reached on this branch
-}
-
 // SyncPath is a path tuple identifying a particular trie node either in a single
 // trie (account) or a layered trie (account -> storage).
 //
@@ -85,30 +73,57 @@ func NewSyncPath(path []byte) SyncPath {
 	return SyncPath{hexToKeybytes(path[:64]), hexToCompact(path[64:])}
 }
 
-// SyncResult is a response with requested data along with it's hash.
-type SyncResult struct {
-	Hash common.Hash // Hash of the originally unknown trie node
-	Data []byte      // Data content of the retrieved node
+// nodeRequest represents a scheduled or already in-flight trie node retrieval request.
+type nodeRequest struct {
+	hash common.Hash // Hash of the trie node to retrieve
+	path []byte      // Merkle path leading to this node for prioritization
+	data []byte      // Data content of the node, cached until all subtrees complete
+
+	parent   *nodeRequest // Parent state node referencing this entry
+	deps     int          // Number of dependencies before allowed to commit this node
+	callback LeafCallback // Callback to invoke if a leaf node it reached on this branch
+}
+
+// codeRequest represents a scheduled or already in-flight bytecode retrieval request.
+type codeRequest struct {
+	hash    common.Hash    // Hash of the contract bytecode to retrieve
+	path    []byte         // Merkle path leading to this node for prioritization
+	data    []byte         // Data content of the node, cached until all subtrees complete
+	parents []*nodeRequest // Parent state nodes referencing this entry (notify all upon completion)
+}
+
+// NodeSyncResult is a response with requested trie node along with its node path.
+type NodeSyncResult struct {
+	Path string // Path of the originally unknown trie node
+	Data []byte // Data content of the retrieved trie node
+}
+
+// CodeSyncResult is a response with requested bytecode along with its hash.
+type CodeSyncResult struct {
+	Hash common.Hash // Hash the originally unknown bytecode
+	Data []byte      // Data content of the retrieved bytecode
 }
 
 // syncMemBatch is an in-memory buffer of successfully downloaded but not yet
 // persisted data items.
 type syncMemBatch struct {
-	nodes map[common.Hash][]byte // In-memory membatch of recently completed nodes
-	codes map[common.Hash][]byte // In-memory membatch of recently completed codes
+	nodes  map[string][]byte      // In-memory membatch of recently completed nodes
+	hashes map[string]common.Hash // Hashes of recently completed nodes
+	codes  map[common.Hash][]byte // In-memory membatch of recently completed codes
 }
 
 // newSyncMemBatch allocates a new memory-buffer for not-yet persisted trie nodes.
 func newSyncMemBatch() *syncMemBatch {
 	return &syncMemBatch{
-		nodes: make(map[common.Hash][]byte),
-		codes: make(map[common.Hash][]byte),
+		nodes:  make(map[string][]byte),
+		hashes: make(map[string]common.Hash),
+		codes:  make(map[common.Hash][]byte),
 	}
 }
 
-// hasNode reports the trie node with specific hash is already cached.
-func (batch *syncMemBatch) hasNode(hash common.Hash) bool {
-	_, ok := batch.nodes[hash]
+// hasNode reports the trie node with specific path is already cached.
+func (batch *syncMemBatch) hasNode(path []byte) bool {
+	_, ok := batch.nodes[string(path)]
 	return ok
 }
 
@@ -122,12 +137,12 @@ func (batch *syncMemBatch) hasCode(hash common.Hash) bool {
 // unknown trie hashes to retrieve, accepts node data associated with said hashes
 // and reconstructs the trie step by step until all is done.
 type Sync struct {
-	database ethdb.KeyValueReader     // Persistent database to check for existing entries
-	membatch *syncMemBatch            // Memory buffer to avoid frequent database writes
-	nodeReqs map[common.Hash]*request // Pending requests pertaining to a trie node hash
-	codeReqs map[common.Hash]*request // Pending requests pertaining to a code hash
-	queue    *prque.Prque             // Priority queue with the pending requests
-	fetches  map[int]int              // Number of active fetches per trie node depth
+	database ethdb.KeyValueReader         // Persistent database to check for existing entries
+	membatch *syncMemBatch                // Memory buffer to avoid frequent database writes
+	nodeReqs map[string]*nodeRequest      // Pending requests pertaining to a trie node path
+	codeReqs map[common.Hash]*codeRequest // Pending requests pertaining to a code hash
+	queue    *prque.Prque                 // Priority queue with the pending requests
+	fetches  map[int]int                  // Number of active fetches per trie node depth
 }
 
 // NewSync creates a new trie data download scheduler.
@@ -135,51 +150,51 @@ func NewSync(root common.Hash, database ethdb.KeyValueReader, callback LeafCallb
 	ts := &Sync{
 		database: database,
 		membatch: newSyncMemBatch(),
-		nodeReqs: make(map[common.Hash]*request),
-		codeReqs: make(map[common.Hash]*request),
+		nodeReqs: make(map[string]*nodeRequest),
+		codeReqs: make(map[common.Hash]*codeRequest),
 		queue:    prque.New(nil),
 		fetches:  make(map[int]int),
 	}
-	ts.AddSubTrie(root, nil, common.Hash{}, callback)
+	ts.AddSubTrie(root, nil, common.Hash{}, nil, callback)
 	return ts
 }
 
-// AddSubTrie registers a new trie to the sync code, rooted at the designated parent.
-func (s *Sync) AddSubTrie(root common.Hash, path []byte, parent common.Hash, callback LeafCallback) {
+// AddSubTrie registers a new trie to the sync code, rooted at the designated
+// parent for completion tracking. The given path is a unique node path in
+// hex format and contain all the parent path if it's layered trie node.
+func (s *Sync) AddSubTrie(root common.Hash, path []byte, parent common.Hash, parentPath []byte, callback LeafCallback) {
 	// Short circuit if the trie is empty or already known
 	if root == emptyRoot {
 		return
 	}
-	if s.membatch.hasNode(root) {
+	if s.membatch.hasNode(path) {
 		return
 	}
-	// If database says this is a duplicate, then at least the trie node is
-	// present, and we hold the assumption that it's NOT legacy contract code.
 	if rawdb.HasTrieNode(s.database, root) {
 		return
 	}
 	// Assemble the new sub-trie sync request
-	req := &request{
-		path:     path,
+	req := &nodeRequest{
 		hash:     root,
+		path:     path,
 		callback: callback,
 	}
 	// If this sub-trie has a designated parent, link them together
 	if parent != (common.Hash{}) {
-		ancestor := s.nodeReqs[parent]
+		ancestor := s.nodeReqs[string(parentPath)]
 		if ancestor == nil {
 			panic(fmt.Sprintf("sub-trie ancestor not found: %x", parent))
 		}
 		ancestor.deps++
-		req.parents = append(req.parents, ancestor)
+		req.parent = ancestor
 	}
-	s.schedule(req)
+	s.scheduleNodeRequest(req)
 }
 
 // AddCodeEntry schedules the direct retrieval of a contract code that should not
 // be interpreted as a trie node, but rather accepted and stored into the database
 // as is.
-func (s *Sync) AddCodeEntry(hash common.Hash, path []byte, parent common.Hash) {
+func (s *Sync) AddCodeEntry(hash common.Hash, path []byte, parent common.Hash, parentPath []byte) {
 	// Short circuit if the entry is empty or already known
 	if hash == emptyState {
 		return
@@ -196,30 +211,29 @@ func (s *Sync) AddCodeEntry(hash common.Hash, path []byte, parent common.Hash) {
 		return
 	}
 	// Assemble the new sub-trie sync request
-	req := &request{
+	req := &codeRequest{
 		path: path,
 		hash: hash,
-		code: true,
 	}
 	// If this sub-trie has a designated parent, link them together
 	if parent != (common.Hash{}) {
-		ancestor := s.nodeReqs[parent] // the parent of codereq can ONLY be nodereq
+		ancestor := s.nodeReqs[string(parentPath)] // the parent of codereq can ONLY be nodereq
 		if ancestor == nil {
 			panic(fmt.Sprintf("raw-entry ancestor not found: %x", parent))
 		}
 		ancestor.deps++
 		req.parents = append(req.parents, ancestor)
 	}
-	s.schedule(req)
+	s.scheduleCodeRequest(req)
 }
 
 // Missing retrieves the known missing nodes from the trie for retrieval. To aid
 // both eth/6x style fast sync and snap/1x style state sync, the paths of trie
 // nodes are returned too, as well as separate hash list for codes.
-func (s *Sync) Missing(max int) (nodes []common.Hash, paths []SyncPath, codes []common.Hash) {
+func (s *Sync) Missing(max int) ([]string, []common.Hash, []common.Hash) {
 	var (
+		nodePaths  []string
 		nodeHashes []common.Hash
-		nodePaths  []SyncPath
 		codeHashes []common.Hash
 	)
 	for !s.queue.Empty() && (max == 0 || len(nodeHashes)+len(codeHashes) < max) {
@@ -235,62 +249,77 @@ func (s *Sync) Missing(max int) (nodes []common.Hash, paths []SyncPath, codes []
 		s.queue.Pop()
 		s.fetches[depth]++
 
-		hash := item.(common.Hash)
-		if req, ok := s.nodeReqs[hash]; ok {
-			nodeHashes = append(nodeHashes, hash)
-			nodePaths = append(nodePaths, NewSyncPath(req.path))
-		} else {
-			codeHashes = append(codeHashes, hash)
+		switch item.(type) {
+		case common.Hash:
+			codeHashes = append(codeHashes, item.(common.Hash))
+		case string:
+			path := item.(string)
+			req, ok := s.nodeReqs[path]
+			if !ok {
+				log.Error("Missing node request", "path", path)
+				continue // System very wrong, shouldn't happen
+			}
+			nodePaths = append(nodePaths, path)
+			nodeHashes = append(nodeHashes, req.hash)
 		}
 	}
-	return nodeHashes, nodePaths, codeHashes
+	return nodePaths, nodeHashes, codeHashes
 }
 
-// Process injects the received data for requested item. Note it can
+// ProcessCode injects the received data for requested item. Note it can
 // happpen that the single response commits two pending requests(e.g.
 // there are two requests one for code and one for node but the hash
 // is same). In this case the second response for the same hash will
 // be treated as "non-requested" item or "already-processed" item but
 // there is no downside.
-func (s *Sync) Process(result SyncResult) error {
-	// If the item was not requested either for code or node, bail out
-	if s.nodeReqs[result.Hash] == nil && s.codeReqs[result.Hash] == nil {
+func (s *Sync) ProcessCode(result CodeSyncResult) error {
+	// If the code was not requested or it's already processed, bail out
+	req := s.codeReqs[result.Hash]
+	if req == nil {
 		return ErrNotRequested
 	}
-	// There is an pending code request for this data, commit directly
-	var filled bool
-	if req := s.codeReqs[result.Hash]; req != nil && req.data == nil {
-		filled = true
-		req.data = result.Data
-		s.commit(req)
+	if req.data != nil {
+		return ErrAlreadyProcessed
 	}
-	// There is an pending node request for this data, fill it.
-	if req := s.nodeReqs[result.Hash]; req != nil && req.data == nil {
-		filled = true
-		// Decode the node data content and update the request
-		node, err := decodeNode(result.Hash[:], result.Data)
-		if err != nil {
-			return err
-		}
-		req.data = result.Data
+	req.data = result.Data
+	return s.commitCodeRequest(req)
+}
 
-		// Create and schedule a request for all the children nodes
-		requests, err := s.children(req, node)
-		if err != nil {
-			return err
-		}
-		if len(requests) == 0 && req.deps == 0 {
-			s.commit(req)
-		} else {
-			req.deps += len(requests)
-			for _, child := range requests {
-				s.schedule(child)
-			}
-		}
+// ProcessNode injects the received data for requested item. Note it can
+// happen that the single response commits two pending requests(e.g.
+// there are two requests one for code and one for node but the hash
+// is same). In this case the second response for the same hash will
+// be treated as "non-requested" item or "already-processed" item but
+// there is no downside.
+func (s *Sync) ProcessNode(result NodeSyncResult) error {
+	// If the trie node was not requested or it's already processed, bail out
+	req := s.nodeReqs[result.Path]
+	if req == nil {
+		return ErrNotRequested
 	}
-	if !filled {
+	if req.data != nil {
 		return ErrAlreadyProcessed
 	}
+	// Decode the node data content and update the request
+	node, err := decodeNode(req.hash.Bytes(), result.Data)
+	if err != nil {
+		return err
+	}
+	req.data = result.Data
+
+	// Create and schedule a request for all the children nodes
+	requests, err := s.children(req, node)
+	if err != nil {
+		return err
+	}
+	if len(requests) == 0 && req.deps == 0 {
+		s.commitNodeRequest(req)
+	} else {
+		req.deps += len(requests)
+		for _, child := range requests {
+			s.scheduleNodeRequest(child)
+		}
+	}
 	return nil
 }
 
@@ -298,11 +327,11 @@ func (s *Sync) Process(result SyncResult) error {
 // storage, returning any occurred error.
 func (s *Sync) Commit(dbw ethdb.Batch) error {
 	// Dump the membatch into a database dbw
-	for key, value := range s.membatch.nodes {
-		rawdb.WriteTrieNode(dbw, key, value)
+	for path, value := range s.membatch.nodes {
+		rawdb.WriteTrieNode(dbw, s.membatch.hashes[path], value)
 	}
-	for key, value := range s.membatch.codes {
-		rawdb.WriteCode(dbw, key, value)
+	for hash, value := range s.membatch.codes {
+		rawdb.WriteCode(dbw, hash, value)
 	}
 	// Drop the membatch data and return
 	s.membatch = newSyncMemBatch()
@@ -317,23 +346,31 @@ func (s *Sync) Pending() int {
 // schedule inserts a new state retrieval request into the fetch queue. If there
 // is already a pending request for this node, the new request will be discarded
 // and only a parent reference added to the old one.
-func (s *Sync) schedule(req *request) {
-	var reqset = s.nodeReqs
-	if req.code {
-		reqset = s.codeReqs
+func (s *Sync) scheduleNodeRequest(req *nodeRequest) {
+	s.nodeReqs[string(req.path)] = req
+
+	// Schedule the request for future retrieval. This queue is shared
+	// by both node requests and code requests.
+	prio := int64(len(req.path)) << 56 // depth >= 128 will never happen, storage leaves will be included in their parents
+	for i := 0; i < 14 && i < len(req.path); i++ {
+		prio |= int64(15-req.path[i]) << (52 - i*4) // 15-nibble => lexicographic order
 	}
+	s.queue.Push(string(req.path), prio)
+}
+
+// schedule inserts a new state retrieval request into the fetch queue. If there
+// is already a pending request for this node, the new request will be discarded
+// and only a parent reference added to the old one.
+func (s *Sync) scheduleCodeRequest(req *codeRequest) {
 	// If we're already requesting this node, add a new reference and stop
-	if old, ok := reqset[req.hash]; ok {
+	if old, ok := s.codeReqs[req.hash]; ok {
 		old.parents = append(old.parents, req.parents...)
 		return
 	}
-	reqset[req.hash] = req
+	s.codeReqs[req.hash] = req
 
 	// Schedule the request for future retrieval. This queue is shared
-	// by both node requests and code requests. It can happen that there
-	// is a trie node and code has same hash. In this case two elements
-	// with same hash and same or different depth will be pushed. But it's
-	// ok the worst case is the second response will be treated as duplicated.
+	// by both node requests and code requests.
 	prio := int64(len(req.path)) << 56 // depth >= 128 will never happen, storage leaves will be included in their parents
 	for i := 0; i < 14 && i < len(req.path); i++ {
 		prio |= int64(15-req.path[i]) << (52 - i*4) // 15-nibble => lexicographic order
@@ -343,7 +380,7 @@ func (s *Sync) schedule(req *request) {
 
 // children retrieves all the missing children of a state trie entry for future
 // retrieval scheduling.
-func (s *Sync) children(req *request, object node) ([]*request, error) {
+func (s *Sync) children(req *nodeRequest, object node) ([]*nodeRequest, error) {
 	// Gather all the children of the node, irrelevant whether known or not
 	type child struct {
 		path []byte
@@ -374,7 +411,7 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
 		panic(fmt.Sprintf("unknown node: %+v", node))
 	}
 	// Iterate over the children, and request all unknown ones
-	requests := make([]*request, 0, len(children))
+	requests := make([]*nodeRequest, 0, len(children))
 	for _, child := range children {
 		// Notify any external watcher of a new key/value node
 		if req.callback != nil {
@@ -386,7 +423,7 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
 					paths = append(paths, hexToKeybytes(child.path[:2*common.HashLength]))
 					paths = append(paths, hexToKeybytes(child.path[2*common.HashLength:]))
 				}
-				if err := req.callback(paths, child.path, node, req.hash); err != nil {
+				if err := req.callback(paths, child.path, node, req.hash, req.path); err != nil {
 					return nil, err
 				}
 			}
@@ -394,20 +431,20 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
 		// If the child references another node, resolve or schedule
 		if node, ok := (child.node).(hashNode); ok {
 			// Try to resolve the node from the local database
-			hash := common.BytesToHash(node)
-			if s.membatch.hasNode(hash) {
+			if s.membatch.hasNode(child.path) {
 				continue
 			}
 			// If database says duplicate, then at least the trie node is present
 			// and we hold the assumption that it's NOT legacy contract code.
-			if rawdb.HasTrieNode(s.database, hash) {
+			chash := common.BytesToHash(node)
+			if rawdb.HasTrieNode(s.database, chash) {
 				continue
 			}
 			// Locally unknown node, schedule for retrieval
-			requests = append(requests, &request{
+			requests = append(requests, &nodeRequest{
 				path:     child.path,
-				hash:     hash,
-				parents:  []*request{req},
+				hash:     chash,
+				parent:   req,
 				callback: req.callback,
 			})
 		}
@@ -418,22 +455,40 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
 // commit finalizes a retrieval request and stores it into the membatch. If any
 // of the referencing parent requests complete due to this commit, they are also
 // committed themselves.
-func (s *Sync) commit(req *request) (err error) {
+func (s *Sync) commitNodeRequest(req *nodeRequest) error {
 	// Write the node content to the membatch
-	if req.code {
-		s.membatch.codes[req.hash] = req.data
-		delete(s.codeReqs, req.hash)
-		s.fetches[len(req.path)]--
-	} else {
-		s.membatch.nodes[req.hash] = req.data
-		delete(s.nodeReqs, req.hash)
-		s.fetches[len(req.path)]--
+	s.membatch.nodes[string(req.path)] = req.data
+	s.membatch.hashes[string(req.path)] = req.hash
+
+	delete(s.nodeReqs, string(req.path))
+	s.fetches[len(req.path)]--
+
+	// Check parent for completion
+	if req.parent != nil {
+		req.parent.deps--
+		if req.parent.deps == 0 {
+			if err := s.commitNodeRequest(req.parent); err != nil {
+				return err
+			}
+		}
 	}
+	return nil
+}
+
+// commit finalizes a retrieval request and stores it into the membatch. If any
+// of the referencing parent requests complete due to this commit, they are also
+// committed themselves.
+func (s *Sync) commitCodeRequest(req *codeRequest) error {
+	// Write the node content to the membatch
+	s.membatch.codes[req.hash] = req.data
+	delete(s.codeReqs, req.hash)
+	s.fetches[len(req.path)]--
+
 	// Check all parents for completion
 	for _, parent := range req.parents {
 		parent.deps--
 		if parent.deps == 0 {
-			if err := s.commit(parent); err != nil {
+			if err := s.commitNodeRequest(parent); err != nil {
 				return err
 			}
 		}

trie/sync_test.go

@@ -87,6 +87,13 @@ func checkTrieConsistency(db *Database, root common.Hash) error {
 	return it.Error()
 }
 
+// trieElement represents the element in the state trie(bytecode or trie node).
+type trieElement struct {
+	path     string
+	hash     common.Hash
+	syncPath SyncPath
+}
+
 // Tests that an empty trie is not scheduled for syncing.
 func TestEmptySync(t *testing.T) {
 	dbA := NewDatabase(memorydb.New())
@@ -96,8 +103,8 @@ func TestEmptySync(t *testing.T) {
 
 	for i, trie := range []*Trie{emptyA, emptyB} {
 		sync := NewSync(trie.Hash(), memorydb.New(), nil)
-		if nodes, paths, codes := sync.Missing(1); len(nodes) != 0 || len(paths) != 0 || len(codes) != 0 {
-			t.Errorf("test %d: content requested for empty trie: %v, %v, %v", i, nodes, paths, codes)
+		if paths, nodes, codes := sync.Missing(1); len(paths) != 0 || len(nodes) != 0 || len(codes) != 0 {
+			t.Errorf("test %d: content requested for empty trie: %v, %v, %v", i, paths, nodes, codes)
 		}
 	}
 }
@@ -118,35 +125,38 @@ func testIterativeSync(t *testing.T, count int, bypath bool) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	nodes, paths, codes := sched.Missing(count)
-	var (
-		hashQueue []common.Hash
-		pathQueue []SyncPath
-	)
-	if !bypath {
-		hashQueue = append(append(hashQueue[:0], nodes...), codes...)
-	} else {
-		hashQueue = append(hashQueue[:0], codes...)
-		pathQueue = append(pathQueue[:0], paths...)
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	paths, nodes, _ := sched.Missing(count)
+	var elements []trieElement
+	for i := 0; i < len(paths); i++ {
+		elements = append(elements, trieElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(paths[i])),
+		})
 	}
-	for len(hashQueue)+len(pathQueue) > 0 {
-		results := make([]SyncResult, len(hashQueue)+len(pathQueue))
-		for i, hash := range hashQueue {
-			data, err := srcDb.Node(hash)
-			if err != nil {
-				t.Fatalf("failed to retrieve node data for hash %x: %v", hash, err)
+	for len(elements) > 0 {
+		results := make([]NodeSyncResult, len(elements))
+		if !bypath {
+			for i, element := range elements {
+				data, err := srcDb.Node(element.hash)
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for hash %x: %v", element.hash, err)
+				}
+				results[i] = NodeSyncResult{element.path, data}
 			}
-			results[i] = SyncResult{hash, data}
-		}
-		for i, path := range pathQueue {
-			data, _, err := srcTrie.TryGetNode(path[0])
-			if err != nil {
-				t.Fatalf("failed to retrieve node data for path %x: %v", path, err)
+		} else {
+			for i, element := range elements {
+				data, _, err := srcTrie.TryGetNode(element.syncPath[len(element.syncPath)-1])
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for path %x: %v", element.path, err)
+				}
+				results[i] = NodeSyncResult{element.path, data}
 			}
-			results[len(hashQueue)+i] = SyncResult{crypto.Keccak256Hash(data), data}
 		}
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -156,12 +166,14 @@ func testIterativeSync(t *testing.T, count int, bypath bool) {
 		}
 		batch.Write()
 
-		nodes, paths, codes = sched.Missing(count)
-		if !bypath {
-			hashQueue = append(append(hashQueue[:0], nodes...), codes...)
-		} else {
-			hashQueue = append(hashQueue[:0], codes...)
-			pathQueue = append(pathQueue[:0], paths...)
+		paths, nodes, _ = sched.Missing(count)
+		elements = elements[:0]
+		for i := 0; i < len(paths); i++ {
+			elements = append(elements, trieElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(paths[i])),
+			})
 		}
 	}
 	// Cross check that the two tries are in sync
@@ -179,21 +191,29 @@ func TestIterativeDelayedSync(t *testing.T) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	nodes, _, codes := sched.Missing(10000)
-	queue := append(append([]common.Hash{}, nodes...), codes...)
-
-	for len(queue) > 0 {
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	paths, nodes, _ := sched.Missing(10000)
+	var elements []trieElement
+	for i := 0; i < len(paths); i++ {
+		elements = append(elements, trieElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(paths[i])),
+		})
+	}
+	for len(elements) > 0 {
 		// Sync only half of the scheduled nodes
-		results := make([]SyncResult, len(queue)/2+1)
-		for i, hash := range queue[:len(results)] {
-			data, err := srcDb.Node(hash)
+		results := make([]NodeSyncResult, len(elements)/2+1)
+		for i, element := range elements[:len(results)] {
+			data, err := srcDb.Node(element.hash)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
+				t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
 			}
-			results[i] = SyncResult{hash, data}
+			results[i] = NodeSyncResult{element.path, data}
 		}
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -203,8 +223,15 @@ func TestIterativeDelayedSync(t *testing.T) {
 		}
 		batch.Write()
 
-		nodes, _, codes = sched.Missing(10000)
-		queue = append(append(queue[len(results):], nodes...), codes...)
+		paths, nodes, _ = sched.Missing(10000)
+		elements = elements[len(results):]
+		for i := 0; i < len(paths); i++ {
+			elements = append(elements, trieElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(paths[i])),
+			})
+		}
 	}
 	// Cross check that the two tries are in sync
 	checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)
@@ -225,24 +252,30 @@ func testIterativeRandomSync(t *testing.T, count int) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	queue := make(map[common.Hash]struct{})
-	nodes, _, codes := sched.Missing(count)
-	for _, hash := range append(nodes, codes...) {
-		queue[hash] = struct{}{}
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	paths, nodes, _ := sched.Missing(count)
+	queue := make(map[string]trieElement)
+	for i, path := range paths {
+		queue[path] = trieElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(paths[i])),
+		}
 	}
 	for len(queue) > 0 {
 		// Fetch all the queued nodes in a random order
-		results := make([]SyncResult, 0, len(queue))
-		for hash := range queue {
-			data, err := srcDb.Node(hash)
+		results := make([]NodeSyncResult, 0, len(queue))
+		for path, element := range queue {
+			data, err := srcDb.Node(element.hash)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
+				t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
 			}
-			results = append(results, SyncResult{hash, data})
+			results = append(results, NodeSyncResult{path, data})
 		}
 		// Feed the retrieved results back and queue new tasks
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -252,10 +285,14 @@ func testIterativeRandomSync(t *testing.T, count int) {
 		}
 		batch.Write()
 
-		queue = make(map[common.Hash]struct{})
-		nodes, _, codes = sched.Missing(count)
-		for _, hash := range append(nodes, codes...) {
-			queue[hash] = struct{}{}
+		paths, nodes, _ = sched.Missing(count)
+		queue = make(map[string]trieElement)
+		for i, path := range paths {
+			queue[path] = trieElement{
+				path:     path,
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(path)),
+			}
 		}
 	}
 	// Cross check that the two tries are in sync
@@ -273,20 +310,26 @@ func TestIterativeRandomDelayedSync(t *testing.T) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	queue := make(map[common.Hash]struct{})
-	nodes, _, codes := sched.Missing(10000)
-	for _, hash := range append(nodes, codes...) {
-		queue[hash] = struct{}{}
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	paths, nodes, _ := sched.Missing(10000)
+	queue := make(map[string]trieElement)
+	for i, path := range paths {
+		queue[path] = trieElement{
+			path:     path,
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(path)),
+		}
 	}
 	for len(queue) > 0 {
 		// Sync only half of the scheduled nodes, even those in random order
-		results := make([]SyncResult, 0, len(queue)/2+1)
-		for hash := range queue {
-			data, err := srcDb.Node(hash)
+		results := make([]NodeSyncResult, 0, len(queue)/2+1)
+		for path, element := range queue {
+			data, err := srcDb.Node(element.hash)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
+				t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
 			}
-			results = append(results, SyncResult{hash, data})
+			results = append(results, NodeSyncResult{path, data})
 
 			if len(results) >= cap(results) {
 				break
@@ -294,7 +337,7 @@ func TestIterativeRandomDelayedSync(t *testing.T) {
 		}
 		// Feed the retrieved results back and queue new tasks
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -304,11 +347,15 @@ func TestIterativeRandomDelayedSync(t *testing.T) {
 		}
 		batch.Write()
 		for _, result := range results {
-			delete(queue, result.Hash)
-		}
-		nodes, _, codes = sched.Missing(10000)
-		for _, hash := range append(nodes, codes...) {
-			queue[hash] = struct{}{}
+			delete(queue, result.Path)
+		}
+		paths, nodes, _ = sched.Missing(10000)
+		for i, path := range paths {
+			queue[path] = trieElement{
+				path:     path,
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(path)),
+			}
 		}
 	}
 	// Cross check that the two tries are in sync
@@ -326,26 +373,35 @@ func TestDuplicateAvoidanceSync(t *testing.T) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	nodes, _, codes := sched.Missing(0)
-	queue := append(append([]common.Hash{}, nodes...), codes...)
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	paths, nodes, _ := sched.Missing(0)
+	var elements []trieElement
+	for i := 0; i < len(paths); i++ {
+		elements = append(elements, trieElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(paths[i])),
+		})
+	}
 	requested := make(map[common.Hash]struct{})
 
-	for len(queue) > 0 {
-		results := make([]SyncResult, len(queue))
-		for i, hash := range queue {
-			data, err := srcDb.Node(hash)
+	for len(elements) > 0 {
+		results := make([]NodeSyncResult, len(elements))
+		for i, element := range elements {
+			data, err := srcDb.Node(element.hash)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
+				t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
 			}
-			if _, ok := requested[hash]; ok {
-				t.Errorf("hash %x already requested once", hash)
+			if _, ok := requested[element.hash]; ok {
+				t.Errorf("hash %x already requested once", element.hash)
 			}
-			requested[hash] = struct{}{}
+			requested[element.hash] = struct{}{}
 
-			results[i] = SyncResult{hash, data}
+			results[i] = NodeSyncResult{element.path, data}
 		}
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -355,8 +411,15 @@ func TestDuplicateAvoidanceSync(t *testing.T) {
 		}
 		batch.Write()
 
-		nodes, _, codes = sched.Missing(0)
-		queue = append(append(queue[:0], nodes...), codes...)
+		paths, nodes, _ = sched.Missing(0)
+		elements = elements[:0]
+		for i := 0; i < len(paths); i++ {
+			elements = append(elements, trieElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(paths[i])),
+			})
+		}
 	}
 	// Cross check that the two tries are in sync
 	checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)
@@ -373,23 +436,34 @@ func TestIncompleteSync(t *testing.T) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	var added []common.Hash
-
-	nodes, _, codes := sched.Missing(1)
-	queue := append(append([]common.Hash{}, nodes...), codes...)
-	for len(queue) > 0 {
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	var (
+		added    []common.Hash
+		elements []trieElement
+		root     = srcTrie.Hash()
+	)
+	paths, nodes, _ := sched.Missing(1)
+	for i := 0; i < len(paths); i++ {
+		elements = append(elements, trieElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(paths[i])),
+		})
+	}
+	for len(elements) > 0 {
 		// Fetch a batch of trie nodes
-		results := make([]SyncResult, len(queue))
-		for i, hash := range queue {
-			data, err := srcDb.Node(hash)
+		results := make([]NodeSyncResult, len(elements))
+		for i, element := range elements {
+			data, err := srcDb.Node(element.hash)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
+				t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
 			}
-			results[i] = SyncResult{hash, data}
+			results[i] = NodeSyncResult{element.path, data}
 		}
 		// Process each of the trie nodes
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -398,27 +472,36 @@ func TestIncompleteSync(t *testing.T) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
+
 		for _, result := range results {
-			added = append(added, result.Hash)
+			hash := crypto.Keccak256Hash(result.Data)
+			if hash != root {
+				added = append(added, hash)
+			}
 			// Check that all known sub-tries in the synced trie are complete
-			if err := checkTrieConsistency(triedb, result.Hash); err != nil {
+			if err := checkTrieConsistency(triedb, hash); err != nil {
 				t.Fatalf("trie inconsistent: %v", err)
 			}
 		}
 		// Fetch the next batch to retrieve
-		nodes, _, codes = sched.Missing(1)
-		queue = append(append(queue[:0], nodes...), codes...)
+		paths, nodes, _ = sched.Missing(1)
+		elements = elements[:0]
+		for i := 0; i < len(paths); i++ {
+			elements = append(elements, trieElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(paths[i])),
+			})
+		}
 	}
 	// Sanity check that removing any node from the database is detected
-	for _, node := range added[1:] {
-		key := node.Bytes()
-		value, _ := diskdb.Get(key)
-
-		diskdb.Delete(key)
-		if err := checkTrieConsistency(triedb, added[0]); err == nil {
-			t.Fatalf("trie inconsistency not caught, missing: %x", key)
+	for _, hash := range added {
+		value, _ := diskdb.Get(hash.Bytes())
+		diskdb.Delete(hash.Bytes())
+		if err := checkTrieConsistency(triedb, root); err == nil {
+			t.Fatalf("trie inconsistency not caught, missing: %x", hash)
 		}
-		diskdb.Put(key, value)
+		diskdb.Put(hash.Bytes(), value)
 	}
 }
 
@@ -433,21 +516,33 @@ func TestSyncOrdering(t *testing.T) {
 	triedb := NewDatabase(diskdb)
 	sched := NewSync(srcTrie.Hash(), diskdb, nil)
 
-	nodes, paths, _ := sched.Missing(1)
-	queue := append([]common.Hash{}, nodes...)
-	reqs := append([]SyncPath{}, paths...)
+	// The code requests are ignored here since there is no code
+	// at the testing trie.
+	var (
+		reqs     []SyncPath
+		elements []trieElement
+	)
+	paths, nodes, _ := sched.Missing(1)
+	for i := 0; i < len(paths); i++ {
+		elements = append(elements, trieElement{
+			path:     paths[i],
+			hash:     nodes[i],
+			syncPath: NewSyncPath([]byte(paths[i])),
+		})
+		reqs = append(reqs, NewSyncPath([]byte(paths[i])))
+	}
 
-	for len(queue) > 0 {
-		results := make([]SyncResult, len(queue))
-		for i, hash := range queue {
-			data, err := srcDb.Node(hash)
+	for len(elements) > 0 {
+		results := make([]NodeSyncResult, len(elements))
+		for i, element := range elements {
+			data, err := srcDb.Node(element.hash)
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
+				t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
 			}
-			results[i] = SyncResult{hash, data}
+			results[i] = NodeSyncResult{element.path, data}
 		}
 		for _, result := range results {
-			if err := sched.Process(result); err != nil {
+			if err := sched.ProcessNode(result); err != nil {
 				t.Fatalf("failed to process result %v", err)
 			}
 		}
@@ -457,9 +552,16 @@ func TestSyncOrdering(t *testing.T) {
 		}
 		batch.Write()
 
-		nodes, paths, _ = sched.Missing(1)
-		queue = append(queue[:0], nodes...)
-		reqs = append(reqs, paths...)
+		paths, nodes, _ = sched.Missing(1)
+		elements = elements[:0]
+		for i := 0; i < len(paths); i++ {
+			elements = append(elements, trieElement{
+				path:     paths[i],
+				hash:     nodes[i],
+				syncPath: NewSyncPath([]byte(paths[i])),
+			})
+			reqs = append(reqs, NewSyncPath([]byte(paths[i])))
+		}
 	}
 	// Cross check that the two tries are in sync
 	checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)

trie/trie.go

@@ -42,18 +42,18 @@ var (
 // LeafCallback is a callback type invoked when a trie operation reaches a leaf
 // node.
 //
-// The paths is a path tuple identifying a particular trie node either in a single
-// trie (account) or a layered trie (account -> storage). Each path in the tuple
+// The keys is a path tuple identifying a particular trie node either in a single
+// trie (account) or a layered trie (account -> storage). Each key in the tuple
 // is in the raw format(32 bytes).
 //
-// The hexpath is a composite hexary path identifying the trie node. All the key
+// The path is a composite hexary path identifying the trie node. All the key
 // bytes are converted to the hexary nibbles and composited with the parent path
 // if the trie node is in a layered trie.
 //
 // It's used by state sync and commit to allow handling external references
 // between account and storage tries. And also it's used in the state healing
 // for extracting the raw states(leaf nodes) with corresponding paths.
-type LeafCallback func(paths [][]byte, hexpath []byte, leaf []byte, parent common.Hash) error
+type LeafCallback func(keys [][]byte, path []byte, leaf []byte, parent common.Hash, parentPath []byte) error
 
 // Trie is a Merkle Patricia Trie.
 // The zero value is an empty trie with no database.

trie/trie_test.go

@@ -625,7 +625,7 @@ func BenchmarkCommitAfterHash(b *testing.B) {
 		benchmarkCommitAfterHash(b, nil)
 	})
 	var a types.StateAccount
-	onleaf := func(paths [][]byte, hexpath []byte, leaf []byte, parent common.Hash) error {
+	onleaf := func(paths [][]byte, hexpath []byte, leaf []byte, parent common.Hash, parentPath []byte) error {
 		rlp.DecodeBytes(leaf, &a)
 		return nil
 	}