From 7e29b0b5b4e5cf7ded9a5a75789de6f8121caec9 Mon Sep 17 00:00:00 2001 From: Péter Szilágyi Date: Mon, 28 Dec 2015 15:20:37 +0200 Subject: core/state, trie: add node iterator, test state/trie sync consistency --- trie/iterator.go | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++-- trie/sync_test.go | 102 +++++++++++++++++++++++++++++++++++++++++++--- 2 files changed, 214 insertions(+), 8 deletions(-) (limited to 'trie') diff --git a/trie/iterator.go b/trie/iterator.go index 5f205e081..e79de2e4e 100644 --- a/trie/iterator.go +++ b/trie/iterator.go @@ -18,22 +18,26 @@ package trie import ( "bytes" + "fmt" "github.com/ethereum/go-ethereum/logger" "github.com/ethereum/go-ethereum/logger/glog" ) +// Iterator is a key-value trie iterator to traverse the data contents. type Iterator struct { trie *Trie - Key []byte - Value []byte + Key []byte // Current data key on which the iterator is positioned on + Value []byte // Current data value on which the iterator is positioned on } +// NewIterator creates a new key-value iterator. func NewIterator(trie *Trie) *Iterator { return &Iterator{trie: trie, Key: nil} } +// Next moves the iterator forward with one key-value entry. func (self *Iterator) Next() bool { isIterStart := false if self.Key == nil { @@ -142,6 +146,116 @@ func (self *Iterator) key(node interface{}) []byte { } return self.key(rn) } - return nil } + +// nodeIteratorState represents the iteration state at one particular node of the +// trie, which can be resumed at a later invocation. +type nodeIteratorState struct { + node node // Trie node being iterated + child int // Child to be processed next +} + +// NodeIterator is an iterator to traverse the trie post-order. +type NodeIterator struct { + trie *Trie // Trie being iterated + stack []*nodeIteratorState // Hierarchy of trie nodes persisting the iteration state + + Node node // Current node being iterated (internal representation) + Leaf bool // Flag whether the current node is a value (data) node + LeafBlob []byte // Data blob contained within a leaf (otherwise nil) +} + +// NewNodeIterator creates an post-order trie iterator. +func NewNodeIterator(trie *Trie) *NodeIterator { + if bytes.Compare(trie.Root(), emptyRoot.Bytes()) == 0 { + return new(NodeIterator) + } + return &NodeIterator{trie: trie} +} + +// Next moves the iterator to the next node, returning whether there are any +// further nodes. +func (it *NodeIterator) Next() bool { + it.step() + return it.retrieve() +} + +// step moves the iterator to the next node of the trie. +func (it *NodeIterator) step() { + // Abort if we reached the end of the iteration + if it.trie == nil { + return + } + // Initialize the iterator if we've just started, or pop off the old node otherwise + if len(it.stack) == 0 { + it.stack = append(it.stack, &nodeIteratorState{node: it.trie.root, child: -1}) + if it.stack[0].node == nil { + panic(fmt.Sprintf("root node missing: %x", it.trie.Root())) + } + } else { + it.stack = it.stack[:len(it.stack)-1] + if len(it.stack) == 0 { + it.trie = nil + return + } + } + // Continue iteration to the next child + for { + parent := it.stack[len(it.stack)-1] + if node, ok := parent.node.(fullNode); ok { + // Full node, traverse all children, then the node itself + if parent.child >= len(node) { + break + } + for parent.child++; parent.child < len(node); parent.child++ { + if current := node[parent.child]; current != nil { + it.stack = append(it.stack, &nodeIteratorState{node: current, child: -1}) + break + } + } + } else if node, ok := parent.node.(shortNode); ok { + // Short node, traverse the pointer singleton child, then the node itself + if parent.child >= 0 { + break + } + parent.child++ + it.stack = append(it.stack, &nodeIteratorState{node: node.Val, child: -1}) + } else if node, ok := parent.node.(hashNode); ok { + // Hash node, resolve the hash child from the database, then the node itself + if parent.child >= 0 { + break + } + parent.child++ + + node, err := it.trie.resolveHash(node, nil, nil) + if err != nil { + panic(err) + } + it.stack = append(it.stack, &nodeIteratorState{node: node, child: -1}) + } else { + break + } + } +} + +// retrieve pulls and caches the current trie node the iterator is traversing. +// In case of a value node, the additional leaf blob is also populated with the +// data contents for external interpretation. +// +// The method returns whether there are any more data left for inspection. +func (it *NodeIterator) retrieve() bool { + // Clear out any previously set values + it.Node, it.Leaf, it.LeafBlob = nil, false, nil + + // If the iteration's done, return no available data + if it.trie == nil { + return false + } + // Otherwise retrieve the current node and resolve leaf accessors + it.Node = it.stack[len(it.stack)-1].node + if value, ok := it.Node.(valueNode); ok { + it.Leaf, it.LeafBlob = true, []byte(value) + } + return true +} diff --git a/trie/sync_test.go b/trie/sync_test.go index 9c036a3a9..94d6edd68 100644 --- a/trie/sync_test.go +++ b/trie/sync_test.go @@ -18,6 +18,7 @@ package trie import ( "bytes" + "fmt" "testing" "github.com/ethereum/go-ethereum/common" @@ -33,6 +34,7 @@ func makeTestTrie() (ethdb.Database, *Trie, map[string][]byte) { // Fill it with some arbitrary data content := make(map[string][]byte) for i := byte(0); i < 255; i++ { + // Map the same data under multiple keys key, val := common.LeftPadBytes([]byte{1, i}, 32), []byte{i} content[string(key)] = val trie.Update(key, val) @@ -40,9 +42,19 @@ func makeTestTrie() (ethdb.Database, *Trie, map[string][]byte) { key, val = common.LeftPadBytes([]byte{2, i}, 32), []byte{i} content[string(key)] = val trie.Update(key, val) + + // Add some other data to inflate th trie + for j := byte(3); j < 13; j++ { + key, val = common.LeftPadBytes([]byte{j, i}, 32), []byte{j, i} + content[string(key)] = val + trie.Update(key, val) + } } trie.Commit() + // Remove any potentially cached data from the test trie creation + globalCache.Clear() + // Return the generated trie return db, trie, content } @@ -50,10 +62,17 @@ func makeTestTrie() (ethdb.Database, *Trie, map[string][]byte) { // checkTrieContents cross references a reconstructed trie with an expected data // content map. func checkTrieContents(t *testing.T, db Database, root []byte, content map[string][]byte) { + // Remove any potentially cached data from the trie synchronisation + globalCache.Clear() + + // Check root availability and trie contents trie, err := New(common.BytesToHash(root), db) if err != nil { t.Fatalf("failed to create trie at %x: %v", root, err) } + if err := checkTrieConsistency(db, common.BytesToHash(root)); err != nil { + t.Fatalf("inconsistent trie at %x: %v", root, err) + } for key, val := range content { if have := trie.Get([]byte(key)); bytes.Compare(have, val) != 0 { t.Errorf("entry %x: content mismatch: have %x, want %x", key, have, val) @@ -61,6 +80,28 @@ func checkTrieContents(t *testing.T, db Database, root []byte, content map[strin } } +// checkTrieConsistency checks that all nodes in a trie and indeed present. +func checkTrieConsistency(db Database, root common.Hash) (failure error) { + // Capture any panics by the iterator + defer func() { + if r := recover(); r != nil { + failure = fmt.Errorf("%v", r) + } + }() + // Remove any potentially cached data from the test trie creation or previous checks + globalCache.Clear() + + // Create and iterate a trie rooted in a subnode + trie, err := New(root, db) + if err != nil { + return + } + it := NewNodeIterator(trie) + for it.Next() { + } + return nil +} + // Tests that an empty trie is not scheduled for syncing. func TestEmptyTrieSync(t *testing.T) { emptyA, _ := New(common.Hash{}, nil) @@ -102,7 +143,7 @@ func testIterativeTrieSync(t *testing.T, batch int) { } queue = append(queue[:0], sched.Missing(batch)...) } - // Cross check that the two tries re in sync + // Cross check that the two tries are in sync checkTrieContents(t, dstDb, srcTrie.Root(), srcData) } @@ -132,7 +173,7 @@ func TestIterativeDelayedTrieSync(t *testing.T) { } queue = append(queue[len(results):], sched.Missing(10000)...) } - // Cross check that the two tries re in sync + // Cross check that the two tries are in sync checkTrieContents(t, dstDb, srcTrie.Root(), srcData) } @@ -173,7 +214,7 @@ func testIterativeRandomTrieSync(t *testing.T, batch int) { queue[hash] = struct{}{} } } - // Cross check that the two tries re in sync + // Cross check that the two tries are in sync checkTrieContents(t, dstDb, srcTrie.Root(), srcData) } @@ -216,7 +257,7 @@ func TestIterativeRandomDelayedTrieSync(t *testing.T) { queue[hash] = struct{}{} } } - // Cross check that the two tries re in sync + // Cross check that the two tries are in sync checkTrieContents(t, dstDb, srcTrie.Root(), srcData) } @@ -252,6 +293,57 @@ func TestDuplicateAvoidanceTrieSync(t *testing.T) { } queue = append(queue[:0], sched.Missing(0)...) } - // Cross check that the two tries re in sync + // Cross check that the two tries are in sync checkTrieContents(t, dstDb, srcTrie.Root(), srcData) } + +// Tests that at any point in time during a sync, only complete sub-tries are in +// the database. +func TestIncompleteTrieSync(t *testing.T) { + // Create a random trie to copy + srcDb, srcTrie, _ := makeTestTrie() + + // Create a destination trie and sync with the scheduler + dstDb, _ := ethdb.NewMemDatabase() + sched := NewTrieSync(common.BytesToHash(srcTrie.Root()), dstDb, nil) + + added := []common.Hash{} + queue := append([]common.Hash{}, sched.Missing(1)...) + for len(queue) > 0 { + // Fetch a batch of trie nodes + results := make([]SyncResult, len(queue)) + for i, hash := range queue { + data, err := srcDb.Get(hash.Bytes()) + if err != nil { + t.Fatalf("failed to retrieve node data for %x: %v", hash, err) + } + results[i] = SyncResult{hash, data} + } + // Process each of the trie nodes + if index, err := sched.Process(results); err != nil { + t.Fatalf("failed to process result #%d: %v", index, err) + } + for _, result := range results { + added = append(added, result.Hash) + } + // Check that all known sub-tries in the synced trie is complete + for _, root := range added { + if err := checkTrieConsistency(dstDb, root); err != nil { + t.Fatalf("trie inconsistent: %v", err) + } + } + // Fetch the next batch to retrieve + queue = append(queue[:0], sched.Missing(1)...) + } + // Sanity check that removing any node from the database is detected + for _, node := range added[1:] { + key := node.Bytes() + value, _ := dstDb.Get(key) + + dstDb.Delete(key) + if err := checkTrieConsistency(dstDb, added[0]); err == nil { + t.Fatalf("trie inconsistency not caught, missing: %x", key) + } + dstDb.Put(key, value) + } +} -- cgit