go-ethereum/trie/sync_test.go

777 lines
25 KiB
Go
Raw Normal View History

// Copyright 2015 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
package trie
import (
"bytes"
"fmt"
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/ethdb/memorydb"
"github.com/ethereum/go-ethereum/trie/trienode"
)
// makeTestTrie create a sample test trie to test node-wise reconstruction.
func makeTestTrie(scheme string) (ethdb.Database, *Database, *StateTrie, map[string][]byte) {
// Create an empty trie
db := rawdb.NewMemoryDatabase()
triedb := newTestDatabase(db, scheme)
trie, _ := NewStateTrie(TrieID(types.EmptyRootHash), triedb)
// 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.MustUpdate(key, val)
key, val = common.LeftPadBytes([]byte{2, i}, 32), []byte{i}
content[string(key)] = val
trie.MustUpdate(key, val)
// Add some other data to inflate the trie
for j := byte(3); j < 13; j++ {
key, val = common.LeftPadBytes([]byte{j, i}, 32), []byte{j, i}
content[string(key)] = val
trie.MustUpdate(key, val)
}
}
root, nodes, _ := trie.Commit(false)
if err := triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
panic(fmt.Errorf("failed to commit db %v", err))
}
if err := triedb.Commit(root, false); err != nil {
panic(err)
}
// Re-create the trie based on the new state
trie, _ = NewStateTrie(TrieID(root), triedb)
return db, triedb, trie, content
}
// checkTrieContents cross references a reconstructed trie with an expected data
// content map.
func checkTrieContents(t *testing.T, db ethdb.Database, scheme string, root []byte, content map[string][]byte) {
// Check root availability and trie contents
ndb := newTestDatabase(db, scheme)
trie, err := NewStateTrie(TrieID(common.BytesToHash(root)), ndb)
if err != nil {
t.Fatalf("failed to create trie at %x: %v", root, err)
}
if err := checkTrieConsistency(db, scheme, common.BytesToHash(root)); err != nil {
t.Fatalf("inconsistent trie at %x: %v", root, err)
}
for key, val := range content {
if have := trie.MustGet([]byte(key)); !bytes.Equal(have, val) {
t.Errorf("entry %x: content mismatch: have %x, want %x", key, have, val)
}
}
}
// checkTrieConsistency checks that all nodes in a trie are indeed present.
func checkTrieConsistency(db ethdb.Database, scheme string, root common.Hash) error {
ndb := newTestDatabase(db, scheme)
trie, err := NewStateTrie(TrieID(root), ndb)
if err != nil {
return nil // Consider a non existent state consistent
}
it := trie.MustNodeIterator(nil)
for it.Next(true) {
}
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 := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
dbB := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
dbC := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.PathScheme)
dbD := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.PathScheme)
emptyA := NewEmpty(dbA)
emptyB, _ := New(TrieID(types.EmptyRootHash), dbB)
emptyC := NewEmpty(dbC)
emptyD, _ := New(TrieID(types.EmptyRootHash), dbD)
for i, trie := range []*Trie{emptyA, emptyB, emptyC, emptyD} {
sync := NewSync(trie.Hash(), memorydb.New(), nil, []*Database{dbA, dbB, dbC, dbD}[i].Scheme())
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)
}
}
}
// 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 TestIterativeSync(t *testing.T) {
testIterativeSync(t, 1, false, rawdb.HashScheme)
testIterativeSync(t, 100, false, rawdb.HashScheme)
testIterativeSync(t, 1, true, rawdb.HashScheme)
testIterativeSync(t, 100, true, rawdb.HashScheme)
testIterativeSync(t, 1, false, rawdb.PathScheme)
testIterativeSync(t, 100, false, rawdb.PathScheme)
testIterativeSync(t, 1, true, rawdb.PathScheme)
testIterativeSync(t, 100, true, rawdb.PathScheme)
}
func testIterativeSync(t *testing.T, count int, bypath bool, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// 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])),
})
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
for len(elements) > 0 {
results := make([]NodeSyncResult, len(elements))
if !bypath {
for i, element := range elements {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, 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}
}
} else {
for i, element := range elements {
data, _, err := srcTrie.GetNode(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}
}
}
for _, result := range results {
if err := sched.ProcessNode(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)
eth/downloader: separate state sync from queue (#14460) * eth/downloader: separate state sync from queue Scheduling of state node downloads hogged the downloader queue lock when new requests were scheduled. This caused timeouts for other requests. With this change, state sync is fully independent of all other downloads and doesn't involve the queue at all. State sync is started and checked on in processContent. This is slightly awkward because processContent doesn't have a select loop. Instead, the queue is closed by an auxiliary goroutine when state sync fails. We tried several alternatives to this but settled on the current approach because it's the least amount of change overall. Handling of the pivot block has changed slightly: the queue previously prevented import of pivot block receipts before the state of the pivot block was available. In this commit, the receipt will be imported before the state. This causes an annoyance where the pivot block is committed as fast block head even when state downloads fail. Stay tuned for more updates in this area ;) * eth/downloader: remove cancelTimeout channel * eth/downloader: retry state requests on timeout * eth/downloader: improve comment * eth/downloader: mark peers idle when state sync is done * eth/downloader: move pivot block splitting to processContent This change also ensures that pivot block receipts aren't imported before the pivot block itself. * eth/downloader: limit state node retries * eth/downloader: improve state node error handling and retry check * eth/downloader: remove maxStateNodeRetries It fails the sync too much. * eth/downloader: remove last use of cancelCh in statesync.go Fixes TestDeliverHeadersHang*Fast and (hopefully) the weird cancellation behaviour at the end of fast sync. * eth/downloader: fix leak in runStateSync * eth/downloader: don't run processFullSyncContent in LightSync mode * eth/downloader: improve comments * eth/downloader: fix vet, megacheck * eth/downloader: remove unrequested tasks anyway * eth/downloader, trie: various polishes around duplicate items This commit explicitly tracks duplicate and unexpected state delieveries done against a trie Sync structure, also adding there to import info logs. The commit moves the db batch used to commit trie changes one level deeper so its flushed after every node insertion. This is needed to avoid a lot of duplicate retrievals caused by inconsistencies between Sync internals and database. A better approach is to track not-yet-written states in trie.Sync and flush on commit, but I'm focuing on correctness first now. The commit fixes a regression around pivot block fail count. The counter previously was reset to 1 if and only if a sync cycle progressed (inserted at least 1 entry to the database). The current code reset it already if a node was delivered, which is not stong enough, because unless it ends up written to disk, an attacker can just loop and attack ad infinitum. The commit also fixes a regression around state deliveries and timeouts. The old downloader tracked if a delivery is stale (none of the deliveries were requestedt), in which case it didn't mark the node idle and did not send further requests, since it signals a past timeout. The current code did mark it idle even on stale deliveries, which eventually caused two requests to be in flight at the same time, making the deliveries always stale and mass duplicating retrievals between multiple peers. * eth/downloader: fix state request leak This commit fixes the hang seen sometimes while doing the state sync. The cause of the hang was a rare combination of events: request state data from peer, peer drops and reconnects almost immediately. This caused a new download task to be assigned to the peer, overwriting the old one still waiting for a timeout, which in turned leaked the requests out, never to be retried. The fix is to ensure that a task assignment moves any pending one back into the retry queue. The commit also fixes a regression with peer dropping due to stalls. The current code considered a peer stalling if they timed out delivering 1 item. However, the downloader never requests only one, the minimum is 2 (attempt to fine tune estimated latency/bandwidth). The fix is simply to drop if a timeout is detected at 2 items. Apart from the above bugfixes, the commit contains some code polishes I made while debugging the hang. * core, eth, trie: support batched trie sync db writes * trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
}
batch.Write()
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
checkTrieContents(t, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), srcData)
}
// Tests that the trie scheduler can correctly reconstruct the state even if only
// partial results are returned, and the others sent only later.
func TestIterativeDelayedSync(t *testing.T) {
testIterativeDelayedSync(t, rawdb.HashScheme)
testIterativeDelayedSync(t, rawdb.PathScheme)
}
func testIterativeDelayedSync(t *testing.T, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// 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])),
})
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
for len(elements) > 0 {
// Sync only half of the scheduled nodes
results := make([]NodeSyncResult, len(elements)/2+1)
for i, element := range elements[:len(results)] {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, element.hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
}
results[i] = NodeSyncResult{element.path, data}
}
for _, result := range results {
if err := sched.ProcessNode(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)
eth/downloader: separate state sync from queue (#14460) * eth/downloader: separate state sync from queue Scheduling of state node downloads hogged the downloader queue lock when new requests were scheduled. This caused timeouts for other requests. With this change, state sync is fully independent of all other downloads and doesn't involve the queue at all. State sync is started and checked on in processContent. This is slightly awkward because processContent doesn't have a select loop. Instead, the queue is closed by an auxiliary goroutine when state sync fails. We tried several alternatives to this but settled on the current approach because it's the least amount of change overall. Handling of the pivot block has changed slightly: the queue previously prevented import of pivot block receipts before the state of the pivot block was available. In this commit, the receipt will be imported before the state. This causes an annoyance where the pivot block is committed as fast block head even when state downloads fail. Stay tuned for more updates in this area ;) * eth/downloader: remove cancelTimeout channel * eth/downloader: retry state requests on timeout * eth/downloader: improve comment * eth/downloader: mark peers idle when state sync is done * eth/downloader: move pivot block splitting to processContent This change also ensures that pivot block receipts aren't imported before the pivot block itself. * eth/downloader: limit state node retries * eth/downloader: improve state node error handling and retry check * eth/downloader: remove maxStateNodeRetries It fails the sync too much. * eth/downloader: remove last use of cancelCh in statesync.go Fixes TestDeliverHeadersHang*Fast and (hopefully) the weird cancellation behaviour at the end of fast sync. * eth/downloader: fix leak in runStateSync * eth/downloader: don't run processFullSyncContent in LightSync mode * eth/downloader: improve comments * eth/downloader: fix vet, megacheck * eth/downloader: remove unrequested tasks anyway * eth/downloader, trie: various polishes around duplicate items This commit explicitly tracks duplicate and unexpected state delieveries done against a trie Sync structure, also adding there to import info logs. The commit moves the db batch used to commit trie changes one level deeper so its flushed after every node insertion. This is needed to avoid a lot of duplicate retrievals caused by inconsistencies between Sync internals and database. A better approach is to track not-yet-written states in trie.Sync and flush on commit, but I'm focuing on correctness first now. The commit fixes a regression around pivot block fail count. The counter previously was reset to 1 if and only if a sync cycle progressed (inserted at least 1 entry to the database). The current code reset it already if a node was delivered, which is not stong enough, because unless it ends up written to disk, an attacker can just loop and attack ad infinitum. The commit also fixes a regression around state deliveries and timeouts. The old downloader tracked if a delivery is stale (none of the deliveries were requestedt), in which case it didn't mark the node idle and did not send further requests, since it signals a past timeout. The current code did mark it idle even on stale deliveries, which eventually caused two requests to be in flight at the same time, making the deliveries always stale and mass duplicating retrievals between multiple peers. * eth/downloader: fix state request leak This commit fixes the hang seen sometimes while doing the state sync. The cause of the hang was a rare combination of events: request state data from peer, peer drops and reconnects almost immediately. This caused a new download task to be assigned to the peer, overwriting the old one still waiting for a timeout, which in turned leaked the requests out, never to be retried. The fix is to ensure that a task assignment moves any pending one back into the retry queue. The commit also fixes a regression with peer dropping due to stalls. The current code considered a peer stalling if they timed out delivering 1 item. However, the downloader never requests only one, the minimum is 2 (attempt to fine tune estimated latency/bandwidth). The fix is simply to drop if a timeout is detected at 2 items. Apart from the above bugfixes, the commit contains some code polishes I made while debugging the hang. * core, eth, trie: support batched trie sync db writes * trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
}
batch.Write()
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, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), srcData)
}
// 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, however in a
// random order.
func TestIterativeRandomSyncIndividual(t *testing.T) {
testIterativeRandomSync(t, 1, rawdb.HashScheme)
testIterativeRandomSync(t, 100, rawdb.HashScheme)
testIterativeRandomSync(t, 1, rawdb.PathScheme)
testIterativeRandomSync(t, 100, rawdb.PathScheme)
}
func testIterativeRandomSync(t *testing.T, count int, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// 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])),
}
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
for len(queue) > 0 {
// Fetch all the queued nodes in a random order
results := make([]NodeSyncResult, 0, len(queue))
for path, element := range queue {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, element.hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
}
results = append(results, NodeSyncResult{path, data})
}
// 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 := diskdb.NewBatch()
if err := sched.Commit(batch); err != nil {
t.Fatalf("failed to commit data: %v", err)
eth/downloader: separate state sync from queue (#14460) * eth/downloader: separate state sync from queue Scheduling of state node downloads hogged the downloader queue lock when new requests were scheduled. This caused timeouts for other requests. With this change, state sync is fully independent of all other downloads and doesn't involve the queue at all. State sync is started and checked on in processContent. This is slightly awkward because processContent doesn't have a select loop. Instead, the queue is closed by an auxiliary goroutine when state sync fails. We tried several alternatives to this but settled on the current approach because it's the least amount of change overall. Handling of the pivot block has changed slightly: the queue previously prevented import of pivot block receipts before the state of the pivot block was available. In this commit, the receipt will be imported before the state. This causes an annoyance where the pivot block is committed as fast block head even when state downloads fail. Stay tuned for more updates in this area ;) * eth/downloader: remove cancelTimeout channel * eth/downloader: retry state requests on timeout * eth/downloader: improve comment * eth/downloader: mark peers idle when state sync is done * eth/downloader: move pivot block splitting to processContent This change also ensures that pivot block receipts aren't imported before the pivot block itself. * eth/downloader: limit state node retries * eth/downloader: improve state node error handling and retry check * eth/downloader: remove maxStateNodeRetries It fails the sync too much. * eth/downloader: remove last use of cancelCh in statesync.go Fixes TestDeliverHeadersHang*Fast and (hopefully) the weird cancellation behaviour at the end of fast sync. * eth/downloader: fix leak in runStateSync * eth/downloader: don't run processFullSyncContent in LightSync mode * eth/downloader: improve comments * eth/downloader: fix vet, megacheck * eth/downloader: remove unrequested tasks anyway * eth/downloader, trie: various polishes around duplicate items This commit explicitly tracks duplicate and unexpected state delieveries done against a trie Sync structure, also adding there to import info logs. The commit moves the db batch used to commit trie changes one level deeper so its flushed after every node insertion. This is needed to avoid a lot of duplicate retrievals caused by inconsistencies between Sync internals and database. A better approach is to track not-yet-written states in trie.Sync and flush on commit, but I'm focuing on correctness first now. The commit fixes a regression around pivot block fail count. The counter previously was reset to 1 if and only if a sync cycle progressed (inserted at least 1 entry to the database). The current code reset it already if a node was delivered, which is not stong enough, because unless it ends up written to disk, an attacker can just loop and attack ad infinitum. The commit also fixes a regression around state deliveries and timeouts. The old downloader tracked if a delivery is stale (none of the deliveries were requestedt), in which case it didn't mark the node idle and did not send further requests, since it signals a past timeout. The current code did mark it idle even on stale deliveries, which eventually caused two requests to be in flight at the same time, making the deliveries always stale and mass duplicating retrievals between multiple peers. * eth/downloader: fix state request leak This commit fixes the hang seen sometimes while doing the state sync. The cause of the hang was a rare combination of events: request state data from peer, peer drops and reconnects almost immediately. This caused a new download task to be assigned to the peer, overwriting the old one still waiting for a timeout, which in turned leaked the requests out, never to be retried. The fix is to ensure that a task assignment moves any pending one back into the retry queue. The commit also fixes a regression with peer dropping due to stalls. The current code considered a peer stalling if they timed out delivering 1 item. However, the downloader never requests only one, the minimum is 2 (attempt to fine tune estimated latency/bandwidth). The fix is simply to drop if a timeout is detected at 2 items. Apart from the above bugfixes, the commit contains some code polishes I made while debugging the hang. * core, eth, trie: support batched trie sync db writes * trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
}
batch.Write()
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
checkTrieContents(t, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), srcData)
}
// Tests that the trie scheduler can correctly reconstruct the state even if only
// partial results are returned (Even those randomly), others sent only later.
func TestIterativeRandomDelayedSync(t *testing.T) {
testIterativeRandomDelayedSync(t, rawdb.HashScheme)
testIterativeRandomDelayedSync(t, rawdb.PathScheme)
}
func testIterativeRandomDelayedSync(t *testing.T, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// 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)),
}
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
for len(queue) > 0 {
// Sync only half of the scheduled nodes, even those in random order
results := make([]NodeSyncResult, 0, len(queue)/2+1)
for path, element := range queue {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, element.hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
}
results = append(results, NodeSyncResult{path, data})
if len(results) >= cap(results) {
break
}
}
// 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 := diskdb.NewBatch()
if err := sched.Commit(batch); err != nil {
t.Fatalf("failed to commit data: %v", err)
eth/downloader: separate state sync from queue (#14460) * eth/downloader: separate state sync from queue Scheduling of state node downloads hogged the downloader queue lock when new requests were scheduled. This caused timeouts for other requests. With this change, state sync is fully independent of all other downloads and doesn't involve the queue at all. State sync is started and checked on in processContent. This is slightly awkward because processContent doesn't have a select loop. Instead, the queue is closed by an auxiliary goroutine when state sync fails. We tried several alternatives to this but settled on the current approach because it's the least amount of change overall. Handling of the pivot block has changed slightly: the queue previously prevented import of pivot block receipts before the state of the pivot block was available. In this commit, the receipt will be imported before the state. This causes an annoyance where the pivot block is committed as fast block head even when state downloads fail. Stay tuned for more updates in this area ;) * eth/downloader: remove cancelTimeout channel * eth/downloader: retry state requests on timeout * eth/downloader: improve comment * eth/downloader: mark peers idle when state sync is done * eth/downloader: move pivot block splitting to processContent This change also ensures that pivot block receipts aren't imported before the pivot block itself. * eth/downloader: limit state node retries * eth/downloader: improve state node error handling and retry check * eth/downloader: remove maxStateNodeRetries It fails the sync too much. * eth/downloader: remove last use of cancelCh in statesync.go Fixes TestDeliverHeadersHang*Fast and (hopefully) the weird cancellation behaviour at the end of fast sync. * eth/downloader: fix leak in runStateSync * eth/downloader: don't run processFullSyncContent in LightSync mode * eth/downloader: improve comments * eth/downloader: fix vet, megacheck * eth/downloader: remove unrequested tasks anyway * eth/downloader, trie: various polishes around duplicate items This commit explicitly tracks duplicate and unexpected state delieveries done against a trie Sync structure, also adding there to import info logs. The commit moves the db batch used to commit trie changes one level deeper so its flushed after every node insertion. This is needed to avoid a lot of duplicate retrievals caused by inconsistencies between Sync internals and database. A better approach is to track not-yet-written states in trie.Sync and flush on commit, but I'm focuing on correctness first now. The commit fixes a regression around pivot block fail count. The counter previously was reset to 1 if and only if a sync cycle progressed (inserted at least 1 entry to the database). The current code reset it already if a node was delivered, which is not stong enough, because unless it ends up written to disk, an attacker can just loop and attack ad infinitum. The commit also fixes a regression around state deliveries and timeouts. The old downloader tracked if a delivery is stale (none of the deliveries were requestedt), in which case it didn't mark the node idle and did not send further requests, since it signals a past timeout. The current code did mark it idle even on stale deliveries, which eventually caused two requests to be in flight at the same time, making the deliveries always stale and mass duplicating retrievals between multiple peers. * eth/downloader: fix state request leak This commit fixes the hang seen sometimes while doing the state sync. The cause of the hang was a rare combination of events: request state data from peer, peer drops and reconnects almost immediately. This caused a new download task to be assigned to the peer, overwriting the old one still waiting for a timeout, which in turned leaked the requests out, never to be retried. The fix is to ensure that a task assignment moves any pending one back into the retry queue. The commit also fixes a regression with peer dropping due to stalls. The current code considered a peer stalling if they timed out delivering 1 item. However, the downloader never requests only one, the minimum is 2 (attempt to fine tune estimated latency/bandwidth). The fix is simply to drop if a timeout is detected at 2 items. Apart from the above bugfixes, the commit contains some code polishes I made while debugging the hang. * core, eth, trie: support batched trie sync db writes * trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
}
batch.Write()
for _, result := range results {
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
checkTrieContents(t, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), srcData)
}
// Tests that a trie sync will not request nodes multiple times, even if they
// have such references.
func TestDuplicateAvoidanceSync(t *testing.T) {
testDuplicateAvoidanceSync(t, rawdb.HashScheme)
testDuplicateAvoidanceSync(t, rawdb.PathScheme)
}
func testDuplicateAvoidanceSync(t *testing.T, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// 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])),
})
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
requested := make(map[common.Hash]struct{})
for len(elements) > 0 {
results := make([]NodeSyncResult, len(elements))
for i, element := range elements {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, element.hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
}
if _, ok := requested[element.hash]; ok {
t.Errorf("hash %x already requested once", element.hash)
}
requested[element.hash] = struct{}{}
results[i] = NodeSyncResult{element.path, data}
}
for _, result := range results {
if err := sched.ProcessNode(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)
eth/downloader: separate state sync from queue (#14460) * eth/downloader: separate state sync from queue Scheduling of state node downloads hogged the downloader queue lock when new requests were scheduled. This caused timeouts for other requests. With this change, state sync is fully independent of all other downloads and doesn't involve the queue at all. State sync is started and checked on in processContent. This is slightly awkward because processContent doesn't have a select loop. Instead, the queue is closed by an auxiliary goroutine when state sync fails. We tried several alternatives to this but settled on the current approach because it's the least amount of change overall. Handling of the pivot block has changed slightly: the queue previously prevented import of pivot block receipts before the state of the pivot block was available. In this commit, the receipt will be imported before the state. This causes an annoyance where the pivot block is committed as fast block head even when state downloads fail. Stay tuned for more updates in this area ;) * eth/downloader: remove cancelTimeout channel * eth/downloader: retry state requests on timeout * eth/downloader: improve comment * eth/downloader: mark peers idle when state sync is done * eth/downloader: move pivot block splitting to processContent This change also ensures that pivot block receipts aren't imported before the pivot block itself. * eth/downloader: limit state node retries * eth/downloader: improve state node error handling and retry check * eth/downloader: remove maxStateNodeRetries It fails the sync too much. * eth/downloader: remove last use of cancelCh in statesync.go Fixes TestDeliverHeadersHang*Fast and (hopefully) the weird cancellation behaviour at the end of fast sync. * eth/downloader: fix leak in runStateSync * eth/downloader: don't run processFullSyncContent in LightSync mode * eth/downloader: improve comments * eth/downloader: fix vet, megacheck * eth/downloader: remove unrequested tasks anyway * eth/downloader, trie: various polishes around duplicate items This commit explicitly tracks duplicate and unexpected state delieveries done against a trie Sync structure, also adding there to import info logs. The commit moves the db batch used to commit trie changes one level deeper so its flushed after every node insertion. This is needed to avoid a lot of duplicate retrievals caused by inconsistencies between Sync internals and database. A better approach is to track not-yet-written states in trie.Sync and flush on commit, but I'm focuing on correctness first now. The commit fixes a regression around pivot block fail count. The counter previously was reset to 1 if and only if a sync cycle progressed (inserted at least 1 entry to the database). The current code reset it already if a node was delivered, which is not stong enough, because unless it ends up written to disk, an attacker can just loop and attack ad infinitum. The commit also fixes a regression around state deliveries and timeouts. The old downloader tracked if a delivery is stale (none of the deliveries were requestedt), in which case it didn't mark the node idle and did not send further requests, since it signals a past timeout. The current code did mark it idle even on stale deliveries, which eventually caused two requests to be in flight at the same time, making the deliveries always stale and mass duplicating retrievals between multiple peers. * eth/downloader: fix state request leak This commit fixes the hang seen sometimes while doing the state sync. The cause of the hang was a rare combination of events: request state data from peer, peer drops and reconnects almost immediately. This caused a new download task to be assigned to the peer, overwriting the old one still waiting for a timeout, which in turned leaked the requests out, never to be retried. The fix is to ensure that a task assignment moves any pending one back into the retry queue. The commit also fixes a regression with peer dropping due to stalls. The current code considered a peer stalling if they timed out delivering 1 item. However, the downloader never requests only one, the minimum is 2 (attempt to fine tune estimated latency/bandwidth). The fix is simply to drop if a timeout is detected at 2 items. Apart from the above bugfixes, the commit contains some code polishes I made while debugging the hang. * core, eth, trie: support batched trie sync db writes * trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
}
batch.Write()
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, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), srcData)
}
// Tests that at any point in time during a sync, only complete sub-tries are in
// the database.
func TestIncompleteSyncHash(t *testing.T) {
testIncompleteSync(t, rawdb.HashScheme)
testIncompleteSync(t, rawdb.PathScheme)
}
func testIncompleteSync(t *testing.T, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, _ := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// The code requests are ignored here since there is no code
// at the testing trie.
var (
addedKeys []string
addedHashes []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])),
})
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
for len(elements) > 0 {
// Fetch a batch of trie nodes
results := make([]NodeSyncResult, len(elements))
for i, element := range elements {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, element.hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
}
results[i] = NodeSyncResult{element.path, data}
}
// Process each of the trie nodes
for _, result := range results {
if err := sched.ProcessNode(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)
eth/downloader: separate state sync from queue (#14460) * eth/downloader: separate state sync from queue Scheduling of state node downloads hogged the downloader queue lock when new requests were scheduled. This caused timeouts for other requests. With this change, state sync is fully independent of all other downloads and doesn't involve the queue at all. State sync is started and checked on in processContent. This is slightly awkward because processContent doesn't have a select loop. Instead, the queue is closed by an auxiliary goroutine when state sync fails. We tried several alternatives to this but settled on the current approach because it's the least amount of change overall. Handling of the pivot block has changed slightly: the queue previously prevented import of pivot block receipts before the state of the pivot block was available. In this commit, the receipt will be imported before the state. This causes an annoyance where the pivot block is committed as fast block head even when state downloads fail. Stay tuned for more updates in this area ;) * eth/downloader: remove cancelTimeout channel * eth/downloader: retry state requests on timeout * eth/downloader: improve comment * eth/downloader: mark peers idle when state sync is done * eth/downloader: move pivot block splitting to processContent This change also ensures that pivot block receipts aren't imported before the pivot block itself. * eth/downloader: limit state node retries * eth/downloader: improve state node error handling and retry check * eth/downloader: remove maxStateNodeRetries It fails the sync too much. * eth/downloader: remove last use of cancelCh in statesync.go Fixes TestDeliverHeadersHang*Fast and (hopefully) the weird cancellation behaviour at the end of fast sync. * eth/downloader: fix leak in runStateSync * eth/downloader: don't run processFullSyncContent in LightSync mode * eth/downloader: improve comments * eth/downloader: fix vet, megacheck * eth/downloader: remove unrequested tasks anyway * eth/downloader, trie: various polishes around duplicate items This commit explicitly tracks duplicate and unexpected state delieveries done against a trie Sync structure, also adding there to import info logs. The commit moves the db batch used to commit trie changes one level deeper so its flushed after every node insertion. This is needed to avoid a lot of duplicate retrievals caused by inconsistencies between Sync internals and database. A better approach is to track not-yet-written states in trie.Sync and flush on commit, but I'm focuing on correctness first now. The commit fixes a regression around pivot block fail count. The counter previously was reset to 1 if and only if a sync cycle progressed (inserted at least 1 entry to the database). The current code reset it already if a node was delivered, which is not stong enough, because unless it ends up written to disk, an attacker can just loop and attack ad infinitum. The commit also fixes a regression around state deliveries and timeouts. The old downloader tracked if a delivery is stale (none of the deliveries were requestedt), in which case it didn't mark the node idle and did not send further requests, since it signals a past timeout. The current code did mark it idle even on stale deliveries, which eventually caused two requests to be in flight at the same time, making the deliveries always stale and mass duplicating retrievals between multiple peers. * eth/downloader: fix state request leak This commit fixes the hang seen sometimes while doing the state sync. The cause of the hang was a rare combination of events: request state data from peer, peer drops and reconnects almost immediately. This caused a new download task to be assigned to the peer, overwriting the old one still waiting for a timeout, which in turned leaked the requests out, never to be retried. The fix is to ensure that a task assignment moves any pending one back into the retry queue. The commit also fixes a regression with peer dropping due to stalls. The current code considered a peer stalling if they timed out delivering 1 item. However, the downloader never requests only one, the minimum is 2 (attempt to fine tune estimated latency/bandwidth). The fix is simply to drop if a timeout is detected at 2 items. Apart from the above bugfixes, the commit contains some code polishes I made while debugging the hang. * core, eth, trie: support batched trie sync db writes * trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
}
batch.Write()
for _, result := range results {
hash := crypto.Keccak256Hash(result.Data)
if hash != root {
addedKeys = append(addedKeys, result.Path)
addedHashes = append(addedHashes, crypto.Keccak256Hash(result.Data))
}
}
// Fetch the next batch to retrieve
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 i, path := range addedKeys {
owner, inner := ResolvePath([]byte(path))
nodeHash := addedHashes[i]
value := rawdb.ReadTrieNode(diskdb, owner, inner, nodeHash, scheme)
rawdb.DeleteTrieNode(diskdb, owner, inner, nodeHash, scheme)
if err := checkTrieConsistency(diskdb, srcDb.Scheme(), root); err == nil {
t.Fatalf("trie inconsistency not caught, missing: %x", path)
}
rawdb.WriteTrieNode(diskdb, owner, inner, nodeHash, value, scheme)
}
}
// Tests that trie nodes get scheduled lexicographically when having the same
// depth.
func TestSyncOrdering(t *testing.T) {
testSyncOrdering(t, rawdb.HashScheme)
testSyncOrdering(t, rawdb.PathScheme)
}
func testSyncOrdering(t *testing.T, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler, tracking the requests
diskdb := rawdb.NewMemoryDatabase()
sched := NewSync(srcTrie.Hash(), diskdb, nil, srcDb.Scheme())
// 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])))
}
reader, err := srcDb.Reader(srcTrie.Hash())
if err != nil {
t.Fatalf("State is not available %x", srcTrie.Hash())
}
for len(elements) > 0 {
results := make([]NodeSyncResult, len(elements))
for i, element := range elements {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, element.hash)
if err != nil {
t.Fatalf("failed to retrieve node data for %x: %v", element.hash, err)
}
results[i] = NodeSyncResult{element.path, data}
}
for _, result := range results {
if err := sched.ProcessNode(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()
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, diskdb, srcDb.Scheme(), 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]))
}
}
}
func syncWith(t *testing.T, root common.Hash, db ethdb.Database, srcDb *Database) {
// Create a destination trie and sync with the scheduler
sched := NewSync(root, db, nil, srcDb.Scheme())
// The code requests are ignored here since there is no code
// at the testing trie.
paths, nodes, _ := sched.Missing(1)
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])),
})
}
reader, err := srcDb.Reader(root)
if err != nil {
t.Fatalf("State is not available %x", root)
}
for len(elements) > 0 {
results := make([]NodeSyncResult, len(elements))
for i, element := range elements {
owner, inner := ResolvePath([]byte(element.path))
data, err := reader.Node(owner, inner, 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}
}
for index, result := range results {
if err := sched.ProcessNode(result); err != nil {
t.Fatalf("failed to process result[%d][%v] data %v %v", index, []byte(result.Path), result.Data, err)
}
}
batch := db.NewBatch()
if err := sched.Commit(batch); err != nil {
t.Fatalf("failed to commit data: %v", err)
}
batch.Write()
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])),
})
}
}
}
// Tests that the syncing target is keeping moving which may overwrite the stale
// states synced in the last cycle.
func TestSyncMovingTarget(t *testing.T) {
testSyncMovingTarget(t, rawdb.HashScheme)
testSyncMovingTarget(t, rawdb.PathScheme)
}
func testSyncMovingTarget(t *testing.T, scheme string) {
// Create a random trie to copy
_, srcDb, srcTrie, srcData := makeTestTrie(scheme)
// Create a destination trie and sync with the scheduler
diskdb := rawdb.NewMemoryDatabase()
syncWith(t, srcTrie.Hash(), diskdb, srcDb)
checkTrieContents(t, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), srcData)
// Push more modifications into the src trie, to see if dest trie can still
// sync with it(overwrite stale states)
var (
preRoot = srcTrie.Hash()
diff = make(map[string][]byte)
)
for i := byte(0); i < 10; i++ {
key, val := randBytes(32), randBytes(32)
srcTrie.MustUpdate(key, val)
diff[string(key)] = val
}
root, nodes, _ := srcTrie.Commit(false)
if err := srcDb.Update(root, preRoot, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
panic(err)
}
if err := srcDb.Commit(root, false); err != nil {
panic(err)
}
preRoot = root
srcTrie, _ = NewStateTrie(TrieID(root), srcDb)
syncWith(t, srcTrie.Hash(), diskdb, srcDb)
checkTrieContents(t, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), diff)
// Revert added modifications from the src trie, to see if dest trie can still
// sync with it(overwrite reverted states)
var reverted = make(map[string][]byte)
for k := range diff {
srcTrie.MustDelete([]byte(k))
reverted[k] = nil
}
for k := range srcData {
val := randBytes(32)
srcTrie.MustUpdate([]byte(k), val)
reverted[k] = val
}
root, nodes, _ = srcTrie.Commit(false)
if err := srcDb.Update(root, preRoot, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
panic(err)
}
if err := srcDb.Commit(root, false); err != nil {
panic(err)
}
srcTrie, _ = NewStateTrie(TrieID(root), srcDb)
syncWith(t, srcTrie.Hash(), diskdb, srcDb)
checkTrieContents(t, diskdb, srcDb.Scheme(), srcTrie.Hash().Bytes(), reverted)
}