gaspersic/go-ethereum
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

core, eth, trie: prepare trie sync for path based operation

Browse Source
This commit is contained in:
Péter Szilágyi
2020-08-28 10:50:37 +03:00
parent 5883afb3ef
commit eeaf191633
6 changed files with 480 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
trie/secure_trie.go
View File

@ -79,6 +79,12 @@ func (t *SecureTrie) TryGet(key []byte) ([]byte, error) {
return t.trie.TryGet(t.hashKey(key))
}
// TryGetNode attempts to retrieve a trie node by compact-encoded path. It is not
// possible to use keybyte-encoding as the path might contain odd nibbles.
func (t *SecureTrie) TryGetNode(path []byte) ([]byte, int, error) {
return t.trie.TryGetNode(path)
}
// Update associates key with value in the trie. Subsequent calls to
// Get will return value. If value has length zero, any existing value
// is deleted from the trie and calls to Get will return nil.

66
trie/sync.go
View File

@ -52,6 +52,39 @@ type request struct {
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).
//
// Content wise the tuple either has 1 element if it addresses a node in a single
// trie or 2 elements if it addresses a node in a stacked trie.
//
// To support aiming arbitrary trie nodes, the path needs to support odd nibble
// lengths. To avoid transferring expanded hex form over the network, the last
// part of the tuple (which needs to index into the middle of a trie) is compact
// encoded. In case of a 2-tuple, the first item is always 32 bytes so that is
// simple binary encoded.
//
// Examples:
// - Path 0x9 -> {0x19}
// - Path 0x99 -> {0x0099}
// - Path 0x01234567890123456789012345678901012345678901234567890123456789019 -> {0x0123456789012345678901234567890101234567890123456789012345678901, 0x19}
// - Path 0x012345678901234567890123456789010123456789012345678901234567890199 -> {0x0123456789012345678901234567890101234567890123456789012345678901, 0x0099}
type SyncPath [][]byte
// newSyncPath converts an expanded trie path from nibble form into a compact
// version that can be sent over the network.
func newSyncPath(path []byte) SyncPath {
// If the hash is from the account trie, append a single item, if it
// is from the a storage trie, append a tuple. Note, the length 64 is
// clashing between account leaf and storage root. It's fine though
// because having a trie node at 64 depth means a hash collision was
// found and we're long dead.
if len(path) < 64 {
return SyncPath{hexToCompact(path)}
}
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
@ -193,10 +226,16 @@ func (s *Sync) AddCodeEntry(hash common.Hash, path []byte, parent common.Hash) {
s.schedule(req)
}
// Missing retrieves the known missing nodes from the trie for retrieval.
func (s *Sync) Missing(max int) []common.Hash {
var requests []common.Hash
for !s.queue.Empty() && (max == 0 || len(requests) < max) {
// 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) {
var (
nodeHashes []common.Hash
nodePaths []SyncPath
codeHashes []common.Hash
)
for !s.queue.Empty() && (max == 0 || len(nodeHashes)+len(codeHashes) < max) {
// Retrieve th enext item in line
item, prio := s.queue.Peek()
@ -208,9 +247,16 @@ func (s *Sync) Missing(max int) []common.Hash {
// Item is allowed to be scheduled, add it to the task list
s.queue.Pop()
s.fetches[depth]++
requests = append(requests, item.(common.Hash))
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)
}
}
return requests
return nodeHashes, nodePaths, codeHashes
}
// Process injects the received data for requested item. Note it can
@ -322,9 +368,13 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
switch node := (object).(type) {
case *shortNode:
key := node.Key
if hasTerm(key) {
key = key[:len(key)-1]
}
children = []child{{
node: node.Val,
path: append(append([]byte(nil), req.path...), node.Key...),
path: append(append([]byte(nil), req.path...), key...),
}}
case *fullNode:
for i := 0; i < 17; i++ {
@ -344,7 +394,7 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
// Notify any external watcher of a new key/value node
if req.callback != nil {
if node, ok := (child.node).(valueNode); ok {
if err := req.callback(req.path, node, req.hash); err != nil {
if err := req.callback(child.path, node, req.hash); err != nil {
return nil, err
}
}

149
trie/sync_test.go
View File

@ -21,14 +21,15 @@ import (
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/ethdb/memorydb"
)
// makeTestTrie create a sample test trie to test node-wise reconstruction.
func makeTestTrie() (*Database, *Trie, map[string][]byte) {
func makeTestTrie() (*Database, *SecureTrie, map[string][]byte) {
// Create an empty trie
triedb := NewDatabase(memorydb.New())
trie, _ := New(common.Hash{}, triedb)
trie, _ := NewSecure(common.Hash{}, triedb)
// Fill it with some arbitrary data
content := make(map[string][]byte)
@ -59,7 +60,7 @@ func makeTestTrie() (*Database, *Trie, map[string][]byte) {
// content map.
func checkTrieContents(t *testing.T, db *Database, root []byte, content map[string][]byte) {
// Check root availability and trie contents
trie, err := New(common.BytesToHash(root), db)
trie, err := NewSecure(common.BytesToHash(root), db)
if err != nil {
t.Fatalf("failed to create trie at %x: %v", root, err)
}
@ -76,7 +77,7 @@ func checkTrieContents(t *testing.T, db *Database, root []byte, content map[stri
// checkTrieConsistency checks that all nodes in a trie are indeed present.
func checkTrieConsistency(db *Database, root common.Hash) error {
// Create and iterate a trie rooted in a subnode
trie, err := New(root, db)
trie, err := NewSecure(root, db)
if err != nil {
return nil // Consider a non existent state consistent
}
@ -94,18 +95,21 @@ func TestEmptySync(t *testing.T) {
emptyB, _ := New(emptyRoot, dbB)
for i, trie := range []*Trie{emptyA, emptyB} {
if req := NewSync(trie.Hash(), memorydb.New(), nil, NewSyncBloom(1, memorydb.New())).Missing(1); len(req) != 0 {
t.Errorf("test %d: content requested for empty trie: %v", i, req)
sync := NewSync(trie.Hash(), memorydb.New(), nil, NewSyncBloom(1, memorydb.New()))
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)
}
}
}
// Tests that given a root hash, a trie can sync iteratively on a single thread,
// requesting retrieval tasks and returning all of them in one go.
func TestIterativeSyncIndividual(t *testing.T) { testIterativeSync(t, 1) }
func TestIterativeSyncBatched(t *testing.T) { testIterativeSync(t, 100) }
func TestIterativeSyncIndividual(t *testing.T) { testIterativeSync(t, 1, false) }
func TestIterativeSyncBatched(t *testing.T) { testIterativeSync(t, 100, false) }
func TestIterativeSyncIndividualByPath(t *testing.T) { testIterativeSync(t, 1, true) }
func TestIterativeSyncBatchedByPath(t *testing.T) { testIterativeSync(t, 100, true) }
func testIterativeSync(t *testing.T, count int) {
func testIterativeSync(t *testing.T, count int, bypath bool) {
// Create a random trie to copy
srcDb, srcTrie, srcData := makeTestTrie()
@ -114,16 +118,33 @@ func testIterativeSync(t *testing.T, count int) {
triedb := NewDatabase(diskdb)
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
queue := append([]common.Hash{}, sched.Missing(count)...)
for len(queue) > 0 {
results := make([]SyncResult, len(queue))
for i, hash := range queue {
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...)
}
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 %x: %v", hash, err)
t.Fatalf("failed to retrieve node data for hash %x: %v", hash, err)
}
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)
}
results[len(hashQueue)+i] = SyncResult{crypto.Keccak256Hash(data), data}
}
for _, result := range results {
if err := sched.Process(result); err != nil {
t.Fatalf("failed to process result %v", err)
@ -134,7 +155,14 @@ func testIterativeSync(t *testing.T, count int) {
t.Fatalf("failed to commit data: %v", err)
}
batch.Write()
queue = append(queue[:0], sched.Missing(count)...)
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...)
}
}
// Cross check that the two tries are in sync
checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)
@ -151,7 +179,9 @@ func TestIterativeDelayedSync(t *testing.T) {
triedb := NewDatabase(diskdb)
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
queue := append([]common.Hash{}, sched.Missing(10000)...)
nodes, _, codes := sched.Missing(10000)
queue := append(append([]common.Hash{}, nodes...), codes...)
for len(queue) > 0 {
// Sync only half of the scheduled nodes
results := make([]SyncResult, len(queue)/2+1)
@ -172,7 +202,9 @@ func TestIterativeDelayedSync(t *testing.T) {
t.Fatalf("failed to commit data: %v", err)
}
batch.Write()
queue = append(queue[len(results):], sched.Missing(10000)...)
nodes, _, codes = sched.Missing(10000)
queue = append(append(queue[len(results):], nodes...), codes...)
}
// Cross check that the two tries are in sync
checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)
@ -194,7 +226,8 @@ func testIterativeRandomSync(t *testing.T, count int) {
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
queue := make(map[common.Hash]struct{})
for _, hash := range sched.Missing(count) {
nodes, _, codes := sched.Missing(count)
for _, hash := range append(nodes, codes...) {
queue[hash] = struct{}{}
}
for len(queue) > 0 {
@ -218,8 +251,10 @@ func testIterativeRandomSync(t *testing.T, count int) {
t.Fatalf("failed to commit data: %v", err)
}
batch.Write()
queue = make(map[common.Hash]struct{})
for _, hash := range sched.Missing(count) {
nodes, _, codes = sched.Missing(count)
for _, hash := range append(nodes, codes...) {
queue[hash] = struct{}{}
}
}
@ -239,7 +274,8 @@ func TestIterativeRandomDelayedSync(t *testing.T) {
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
queue := make(map[common.Hash]struct{})
for _, hash := range sched.Missing(10000) {
nodes, _, codes := sched.Missing(10000)
for _, hash := range append(nodes, codes...) {
queue[hash] = struct{}{}
}
for len(queue) > 0 {
@ -270,7 +306,8 @@ func TestIterativeRandomDelayedSync(t *testing.T) {
for _, result := range results {
delete(queue, result.Hash)
}
for _, hash := range sched.Missing(10000) {
nodes, _, codes = sched.Missing(10000)
for _, hash := range append(nodes, codes...) {
queue[hash] = struct{}{}
}
}
@ -289,7 +326,8 @@ func TestDuplicateAvoidanceSync(t *testing.T) {
triedb := NewDatabase(diskdb)
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
queue := append([]common.Hash{}, sched.Missing(0)...)
nodes, _, codes := sched.Missing(0)
queue := append(append([]common.Hash{}, nodes...), codes...)
requested := make(map[common.Hash]struct{})
for len(queue) > 0 {
@ -316,7 +354,9 @@ func TestDuplicateAvoidanceSync(t *testing.T) {
t.Fatalf("failed to commit data: %v", err)
}
batch.Write()
queue = append(queue[:0], sched.Missing(0)...)
nodes, _, codes = sched.Missing(0)
queue = append(append(queue[:0], nodes...), codes...)
}
// Cross check that the two tries are in sync
checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)
@ -334,7 +374,10 @@ func TestIncompleteSync(t *testing.T) {
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
var added []common.Hash
queue := append([]common.Hash{}, sched.Missing(1)...)
nodes, _, codes := sched.Missing(1)
queue := append(append([]common.Hash{}, nodes...), codes...)
for len(queue) > 0 {
// Fetch a batch of trie nodes
results := make([]SyncResult, len(queue))
@ -366,7 +409,8 @@ func TestIncompleteSync(t *testing.T) {
}
}
// Fetch the next batch to retrieve
queue = append(queue[:0], sched.Missing(1)...)
nodes, _, codes = sched.Missing(1)
queue = append(append(queue[:0], nodes...), codes...)
}
// Sanity check that removing any node from the database is detected
for _, node := range added[1:] {
@ -380,3 +424,58 @@ func TestIncompleteSync(t *testing.T) {
diskdb.Put(key, value)
}
}
// Tests that trie nodes get scheduled lexicographically when having the same
// depth.
func TestSyncOrdering(t *testing.T) {
// Create a random trie to copy
srcDb, srcTrie, srcData := makeTestTrie()
// Create a destination trie and sync with the scheduler, tracking the requests
diskdb := memorydb.New()
triedb := NewDatabase(diskdb)
sched := NewSync(srcTrie.Hash(), diskdb, nil, NewSyncBloom(1, diskdb))
nodes, paths, _ := sched.Missing(1)
queue := append([]common.Hash{}, nodes...)
reqs := append([]SyncPath{}, paths...)
for len(queue) > 0 {
results := make([]SyncResult, len(queue))
for i, hash := range queue {
data, err := srcDb.Node(hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", hash, err)
}
results[i] = SyncResult{hash, data}
}
for _, result := range results {
if err := sched.Process(result); err != nil {
t.Fatalf("failed to process result %v", err)
}
}
batch := diskdb.NewBatch()
if err := sched.Commit(batch); err != nil {
t.Fatalf("failed to commit data: %v", err)
}
batch.Write()
nodes, paths, _ = sched.Missing(1)
queue = append(queue[:0], nodes...)
reqs = append(reqs, paths...)
}
// Cross check that the two tries are in sync
checkTrieContents(t, triedb, srcTrie.Hash().Bytes(), srcData)
// Check that the trie nodes have been requested path-ordered
for i := 0; i < len(reqs)-1; i++ {
if len(reqs[i]) > 1 || len(reqs[i+1]) > 1 {
// In the case of the trie tests, there's no storage so the tuples
// must always be single items. 2-tuples should be tested in state.
t.Errorf("Invalid request tuples: len(%v) or len(%v) > 1", reqs[i], reqs[i+1])
}
if bytes.Compare(compactToHex(reqs[i][0]), compactToHex(reqs[i+1][0])) > 0 {
t.Errorf("Invalid request order: %v before %v", compactToHex(reqs[i][0]), compactToHex(reqs[i+1][0]))
}
}
}

84
trie/trie.go
View File

@ -25,6 +25,7 @@ import (
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/rlp"
)
var (
@ -102,8 +103,7 @@ func (t *Trie) Get(key []byte) []byte {
// The value bytes must not be modified by the caller.
// If a node was not found in the database, a MissingNodeError is returned.
func (t *Trie) TryGet(key []byte) ([]byte, error) {
key = keybytesToHex(key)
value, newroot, didResolve, err := t.tryGet(t.root, key, 0)
value, newroot, didResolve, err := t.tryGet(t.root, keybytesToHex(key), 0)
if err == nil && didResolve {
t.root = newroot
}
@ -146,6 +146,86 @@ func (t *Trie) tryGet(origNode node, key []byte, pos int) (value []byte, newnode
}
}
// TryGetNode attempts to retrieve a trie node by compact-encoded path. It is not
// possible to use keybyte-encoding as the path might contain odd nibbles.
func (t *Trie) TryGetNode(path []byte) ([]byte, int, error) {
item, newroot, resolved, err := t.tryGetNode(t.root, compactToHex(path), 0)
if err != nil {
return nil, resolved, err
}
if resolved > 0 {
t.root = newroot
}
if item == nil {
return nil, resolved, nil
}
enc, err := rlp.EncodeToBytes(item)
if err != nil {
log.Error("Encoding existing trie node failed", "err", err)
return nil, resolved, err
}
return enc, resolved, err
}
func (t *Trie) tryGetNode(origNode node, path []byte, pos int) (item node, newnode node, resolved int, err error) {
// If we reached the requested path, return the current node
if pos >= len(path) {
// Don't return collapsed hash nodes though
if _, ok := origNode.(hashNode); !ok {
// Short nodes have expanded keys, compact them before returning
item := origNode
if sn, ok := item.(*shortNode); ok {
item = &shortNode{
Key: hexToCompact(sn.Key),
Val: sn.Val,
}
}
return item, origNode, 0, nil
}
}
// Path still needs to be traversed, descend into children
switch n := (origNode).(type) {
case nil:
// Non-existent path requested, abort
return nil, nil, 0, nil
case valueNode:
// Path prematurely ended, abort
return nil, nil, 0, nil
case *shortNode:
if len(path)-pos < len(n.Key) || !bytes.Equal(n.Key, path[pos:pos+len(n.Key)]) {
// Path branches off from short node
return nil, n, 0, nil
}
item, newnode, resolved, err = t.tryGetNode(n.Val, path, pos+len(n.Key))
if err == nil && resolved > 0 {
n = n.copy()
n.Val = newnode
}
return item, n, resolved, err
case *fullNode:
item, newnode, resolved, err = t.tryGetNode(n.Children[path[pos]], path, pos+1)
if err == nil && resolved > 0 {
n = n.copy()
n.Children[path[pos]] = newnode
}
return item, n, resolved, err
case hashNode:
child, err := t.resolveHash(n, path[:pos])
if err != nil {
return nil, n, 1, err
}
item, newnode, resolved, err := t.tryGetNode(child, path, pos)
return item, newnode, resolved + 1, err
default:
panic(fmt.Sprintf("%T: invalid node: %v", origNode, origNode))
}
}
// Update associates key with value in the trie. Subsequent calls to
// Get will return value. If value has length zero, any existing value
// is deleted from the trie and calls to Get will return nil.