triedb/pathdb: add recovery mechanism in state indexer (ethereum#32447)

Alternative of ethereum#32335, enhancing the history indexer recovery after
unclean shutdown.

Author: rjl493456442

triedb/pathdb/database_test.go

@@ -121,6 +121,8 @@ func (ctx *genctx) storageOriginSet(rawStorageKey bool, t *tester) map[common.Ad
 type tester struct {
 	db        *Database
 	roots     []common.Hash
+	nodes     []*trienode.MergedNodeSet
+	states    []*StateSetWithOrigin
 	preimages map[common.Hash][]byte
 
 	// current state set
@@ -135,12 +137,38 @@ type tester struct {
 	snapNodes map[common.Hash]*trienode.MergedNodeSet
 }
 
+// testerConfig holds configuration parameters for running a test scenario.
 type testerConfig struct {
-	stateHistory uint64
-	isVerkle     bool
-	layers       int
-	enableIndex  bool
-	journalDir   string
+	stateHistory uint64 // Number of historical states to retain
+	layers       int    // Number of state transitions to generate for
+	enableIndex  bool   // Enable state history indexing or not
+	journalDir   string // Directory path for persisting journal files
+	isVerkle     bool   // Enables Verkle trie mode if true
+
+	writeBuffer *int // Optional, the size of memory allocated for write buffer
+	trieCache   *int // Optional, the size of memory allocated for trie cache
+	stateCache  *int // Optional, the size of memory allocated for state cache
+}
+
+func (c *testerConfig) trieCacheSize() int {
+	if c.trieCache != nil {
+		return *c.trieCache
+	}
+	return 256 * 1024
+}
+
+func (c *testerConfig) stateCacheSize() int {
+	if c.stateCache != nil {
+		return *c.stateCache
+	}
+	return 256 * 1024
+}
+
+func (c *testerConfig) writeBufferSize() int {
+	if c.writeBuffer != nil {
+		return *c.writeBuffer
+	}
+	return 256 * 1024
 }
 
 func newTester(t *testing.T, config *testerConfig) *tester {
@@ -149,9 +177,9 @@ func newTester(t *testing.T, config *testerConfig) *tester {
 		db      = New(disk, &Config{
 			StateHistory:        config.stateHistory,
 			EnableStateIndexing: config.enableIndex,
-			TrieCleanSize:       256 * 1024,
-			StateCleanSize:      256 * 1024,
-			WriteBufferSize:     256 * 1024,
+			TrieCleanSize:       config.trieCacheSize(),
+			StateCleanSize:      config.stateCacheSize(),
+			WriteBufferSize:     config.writeBufferSize(),
 			NoAsyncFlush:        true,
 			JournalDirectory:    config.journalDir,
 		}, config.isVerkle)
@@ -177,6 +205,8 @@ func newTester(t *testing.T, config *testerConfig) *tester {
 			panic(fmt.Errorf("failed to update state changes, err: %w", err))
 		}
 		obj.roots = append(obj.roots, root)
+		obj.nodes = append(obj.nodes, nodes)
+		obj.states = append(obj.states, states)
 	}
 	return obj
 }
@@ -200,6 +230,8 @@ func (t *tester) extend(layers int) {
 			panic(fmt.Errorf("failed to update state changes, err: %w", err))
 		}
 		t.roots = append(t.roots, root)
+		t.nodes = append(t.nodes, nodes)
+		t.states = append(t.states, states)
 	}
 }
 
@@ -885,3 +917,107 @@ func copyStorages(set map[common.Hash]map[common.Hash][]byte) map[common.Hash]ma
 	}
 	return copied
 }
+
+func TestDatabaseIndexRecovery(t *testing.T) {
+	maxDiffLayers = 4
+	defer func() {
+		maxDiffLayers = 128
+	}()
+
+	//log.SetDefault(log.NewLogger(log.NewTerminalHandlerWithLevel(os.Stderr, log.LevelDebug, true)))
+	writeBuffer := 512 * 1024
+	config := &testerConfig{
+		layers:      64,
+		enableIndex: true,
+		writeBuffer: &writeBuffer,
+	}
+	env := newTester(t, config)
+	defer env.release()
+
+	// Ensure the buffer in disk layer is not empty
+	var (
+		bRoot = env.db.tree.bottom().rootHash()
+		dRoot = crypto.Keccak256Hash(rawdb.ReadAccountTrieNode(env.db.diskdb, nil))
+	)
+	for dRoot == bRoot {
+		env.extend(1)
+
+		bRoot = env.db.tree.bottom().rootHash()
+		dRoot = crypto.Keccak256Hash(rawdb.ReadAccountTrieNode(env.db.diskdb, nil))
+	}
+	waitIndexing(env.db)
+
+	var (
+		dIndex int
+		roots  = env.roots
+		hr     = newHistoryReader(env.db.diskdb, env.db.stateFreezer)
+	)
+	for i, root := range roots {
+		if root == dRoot {
+			dIndex = i
+		}
+		if root == bRoot {
+			break
+		}
+		if err := checkHistoricalState(env, root, uint64(i+1), hr); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	// Terminate the database and mutate the journal, it's for simulating
+	// the unclean shutdown
+	env.db.Journal(env.lastHash())
+	env.db.Close()
+
+	// Mutate the journal in disk, it should be regarded as invalid
+	blob := rawdb.ReadTrieJournal(env.db.diskdb)
+	blob[0] = 0xa
+	rawdb.WriteTrieJournal(env.db.diskdb, blob)
+
+	// Reload the database, the extra state histories should be removed
+	env.db = New(env.db.diskdb, env.db.config, false)
+
+	for i := range roots {
+		_, err := readStateHistory(env.db.stateFreezer, uint64(i+1))
+		if i <= dIndex && err != nil {
+			t.Fatalf("State history is not found, %d", i)
+		}
+		if i > dIndex && err == nil {
+			t.Fatalf("Unexpected state history found, %d", i)
+		}
+	}
+	remain, err := env.db.IndexProgress()
+	if err != nil {
+		t.Fatalf("Failed to obtain the progress, %v", err)
+	}
+	if remain == 0 {
+		t.Fatalf("Unexpected progress remain, %d", remain)
+	}
+
+	// Apply new states on top, ensuring state indexing can respond correctly
+	for i := dIndex + 1; i < len(roots); i++ {
+		if err := env.db.Update(roots[i], roots[i-1], uint64(i), env.nodes[i], env.states[i]); err != nil {
+			panic(fmt.Errorf("failed to update state changes, err: %w", err))
+		}
+	}
+	remain, err = env.db.IndexProgress()
+	if err != nil {
+		t.Fatalf("Failed to obtain the progress, %v", err)
+	}
+	if remain != 0 {
+		t.Fatalf("Unexpected progress remain, %d", remain)
+	}
+	waitIndexing(env.db)
+
+	// Ensure the truncated state histories become accessible
+	bRoot = env.db.tree.bottom().rootHash()
+	hr = newHistoryReader(env.db.diskdb, env.db.stateFreezer)
+	for i, root := range roots {
+		if root == bRoot {
+			break
+		}
+		if err := checkHistoricalState(env, root, uint64(i+1), hr); err != nil {
+			t.Fatal(err)
+		}
+	}
+}

triedb/pathdb/history_indexer.go

@@ -322,15 +322,22 @@ func newIndexIniter(disk ethdb.KeyValueStore, freezer ethdb.AncientStore, lastID
 		closed:    make(chan struct{}),
 	}
 	// Load indexing progress
+	var recover bool
 	initer.last.Store(lastID)
 	metadata := loadIndexMetadata(disk)
 	if metadata != nil {
 		initer.indexed.Store(metadata.Last)
+		recover = metadata.Last > lastID
 	}
 
 	// Launch background indexer
 	initer.wg.Add(1)
-	go initer.run(lastID)
+	if recover {
+		log.Info("History indexer is recovering", "history", lastID, "indexed", metadata.Last)
+		go initer.recover(lastID)
+	} else {
+		go initer.run(lastID)
+	}
 	return initer
 }
 
@@ -364,8 +371,8 @@ func (i *indexIniter) remain() uint64 {
 	default:
 		last, indexed := i.last.Load(), i.indexed.Load()
 		if last < indexed {
-			log.Error("Invalid state indexing range", "last", last, "indexed", indexed)
-			return 0
+			log.Warn("State indexer is in recovery", "indexed", indexed, "last", last)
+			return indexed - last
 		}
 		return last - indexed
 	}
@@ -569,6 +576,49 @@ func (i *indexIniter) index(done chan struct{}, interrupt *atomic.Int32, lastID
 	log.Info("Indexed state history", "from", beginID, "to", lastID, "elapsed", common.PrettyDuration(time.Since(start)))
 }
 
+// recover handles unclean shutdown recovery. After an unclean shutdown, any
+// extra histories are typically truncated, while the corresponding history index
+// entries may still have been written. Ideally, we would unindex these histories
+// in reverse order, but there is no guarantee that the required histories will
+// still be available.
+//
+// As a workaround, indexIniter waits until the missing histories are regenerated
+// by chain recovery, under the assumption that the recovered histories will be
+// identical to the lost ones. Fork-awareness should be added in the future to
+// correctly handle histories affected by reorgs.
+func (i *indexIniter) recover(lastID uint64) {
+	defer i.wg.Done()
+
+	for {
+		select {
+		case signal := <-i.interrupt:
+			newLastID := signal.newLastID
+			if newLastID != lastID+1 && newLastID != lastID-1 {
+				signal.result <- fmt.Errorf("invalid history id, last: %d, got: %d", lastID, newLastID)
+				continue
+			}
+
+			// Update the last indexed flag
+			lastID = newLastID
+			signal.result <- nil
+			i.last.Store(newLastID)
+			log.Debug("Updated history index flag", "last", lastID)
+
+			// Terminate the recovery routine once the histories are fully aligned
+			// with the index data, indicating that index initialization is complete.
+			metadata := loadIndexMetadata(i.disk)
+			if metadata != nil && metadata.Last == lastID {
+				close(i.done)
+				log.Info("History indexer is recovered", "last", lastID)
+				return
+			}
+
+		case <-i.closed:
+			return
+		}
+	}
+}
+
 // historyIndexer manages the indexing and unindexing of state histories,
 // providing access to historical states.
 //

triedb/pathdb/history_reader_test.go

@@ -144,7 +144,13 @@ func testHistoryReader(t *testing.T, historyLimit uint64) {
 		maxDiffLayers = 128
 	}()
 
-	env := newTester(t, &testerConfig{stateHistory: historyLimit, layers: 64, enableIndex: true})
+	//log.SetDefault(log.NewLogger(log.NewTerminalHandlerWithLevel(os.Stderr, log.LevelDebug, true)))
+	config := &testerConfig{
+		stateHistory: historyLimit,
+		layers:       64,
+		enableIndex:  true,
+	}
+	env := newTester(t, config)
 	defer env.release()
 	waitIndexing(env.db)
 
@@ -183,7 +189,11 @@ func TestHistoricalStateReader(t *testing.T) {
 	}()
 
 	//log.SetDefault(log.NewLogger(log.NewTerminalHandlerWithLevel(os.Stderr, log.LevelDebug, true)))
-	config := &testerConfig{stateHistory: 0, layers: 64, enableIndex: true}
+	config := &testerConfig{
+		stateHistory: 0,
+		layers:       64,
+		enableIndex:  true,
+	}
 	env := newTester(t, config)
 	defer env.release()
 	waitIndexing(env.db)

triedb/pathdb/journal.go

@@ -267,7 +267,7 @@ func (dl *diskLayer) journal(w io.Writer) error {
 	if err := dl.buffer.states.encode(w); err != nil {
 		return err
 	}
-	log.Debug("Journaled pathdb disk layer", "root", dl.root)
+	log.Debug("Journaled pathdb disk layer", "root", dl.root, "id", dl.id)
 	return nil
 }