core/state/snapshot: unlink snapshots from blocks, quad->linear cleanup

core/blockchain.go

@@ -140,10 +140,10 @@ type BlockChain struct {
 	chainConfig *params.ChainConfig // Chain & network configuration
 	cacheConfig *CacheConfig        // Cache configuration for pruning
 
-	db     ethdb.Database         // Low level persistent database to store final content in
-	snaps  *snapshot.SnapshotTree // Snapshot tree for fast trie leaf access
-	triegc *prque.Prque           // Priority queue mapping block numbers to tries to gc
-	gcproc time.Duration          // Accumulates canonical block processing for trie dumping
+	db     ethdb.Database // Low level persistent database to store final content in
+	snaps  *snapshot.Tree // Snapshot tree for fast trie leaf access
+	triegc *prque.Prque   // Priority queue mapping block numbers to tries to gc
+	gcproc time.Duration  // Accumulates canonical block processing for trie dumping
 
 	hc            *HeaderChain
 	rmLogsFeed    event.Feed
@@ -301,7 +301,7 @@ func NewBlockChain(db ethdb.Database, cacheConfig *CacheConfig, chainConfig *par
 	}
 	// Load any existing snapshot, regenerating it if loading failed
 	head := bc.CurrentBlock()
-	if bc.snaps, err = snapshot.New(bc.db, "snapshot.rlp", head.NumberU64(), head.Root()); err != nil {
+	if bc.snaps, err = snapshot.New(bc.db, "snapshot.rlp", head.Root()); err != nil {
 		return nil, err
 	}
 	// Take ownership of this particular state

core/rawdb/accessors_snapshot.go

@@ -17,38 +17,36 @@
 package rawdb
 
 import (
-	"encoding/binary"
-
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
 )
 
-// ReadSnapshotBlock retrieves the number and root of the block whose state is
-// contained in the persisted snapshot.
-func ReadSnapshotBlock(db ethdb.KeyValueReader) (uint64, common.Hash) {
-	data, _ := db.Get(snapshotBlockKey)
-	if len(data) != 8+common.HashLength {
-		return 0, common.Hash{}
+// ReadSnapshotRoot retrieves the root of the block whose state is contained in
+// the persisted snapshot.
+func ReadSnapshotRoot(db ethdb.KeyValueReader) common.Hash {
+	data, _ := db.Get(snapshotRootKey)
+	if len(data) != common.HashLength {
+		return common.Hash{}
 	}
-	return binary.BigEndian.Uint64(data[:8]), common.BytesToHash(data[8:])
+	return common.BytesToHash(data)
 }
 
-// WriteSnapshotBlock stores the number and root of the block whose state is
-// contained in the persisted snapshot.
-func WriteSnapshotBlock(db ethdb.KeyValueWriter, number uint64, root common.Hash) {
-	if err := db.Put(snapshotBlockKey, append(encodeBlockNumber(number), root.Bytes()...)); err != nil {
-		log.Crit("Failed to store snapsnot block's number and root", "err", err)
+// WriteSnapshotRoot stores the root of the block whose state is contained in
+// the persisted snapshot.
+func WriteSnapshotRoot(db ethdb.KeyValueWriter, root common.Hash) {
+	if err := db.Put(snapshotRootKey, root[:]); err != nil {
+		log.Crit("Failed to store snapshot root", "err", err)
 	}
 }
 
-// DeleteSnapshotBlock deletes the number and hash of the block whose state is
-// contained in the persisted snapshot. Since snapshots are not immutable, this
-// method can be used during updates, so a crash or failure will mark the entire
-// snapshot invalid.
-func DeleteSnapshotBlock(db ethdb.KeyValueWriter) {
-	if err := db.Delete(snapshotBlockKey); err != nil {
-		log.Crit("Failed to remove snapsnot block's number and hash", "err", err)
+// DeleteSnapshotRoot deletes the hash of the block whose state is contained in
+// the persisted snapshot. Since snapshots are not immutable, this  method can
+// be used during updates, so a crash or failure will mark the entire snapshot
+// invalid.
+func DeleteSnapshotRoot(db ethdb.KeyValueWriter) {
+	if err := db.Delete(snapshotRootKey); err != nil {
+		log.Crit("Failed to remove snapshot root", "err", err)
 	}
 }
 

core/rawdb/schema.go

@@ -41,8 +41,8 @@ var (
 	// fastTrieProgressKey tracks the number of trie entries imported during fast sync.
 	fastTrieProgressKey = []byte("TrieSync")
 
-	// snapshotBlockKey tracks the number and hash of the last snapshot.
-	snapshotBlockKey = []byte("SnapshotBlock")
+	// snapshotRootKey tracks the number and hash of the last snapshot.
+	snapshotRootKey = []byte("SnapshotRoot")
 
 	// Data item prefixes (use single byte to avoid mixing data types, avoid `i`, used for indexes).
 	headerPrefix       = []byte("h") // headerPrefix + num (uint64 big endian) + hash -> header

core/state/snapshot/difflayer.go

@@ -36,9 +36,8 @@ type diffLayer struct {
 	parent snapshot // Parent snapshot modified by this one, never nil
 	memory uint64   // Approximate guess as to how much memory we use
 
-	number uint64      // Block number to which this snapshot diff belongs to
-	root   common.Hash // Root hash to which this snapshot diff belongs to
-	stale  bool        // Signals that the layer became stale (state progressed)
+	root  common.Hash // Root hash to which this snapshot diff belongs to
+	stale bool        // Signals that the layer became stale (state progressed)
 
 	accountList []common.Hash                          // List of account for iteration. If it exists, it's sorted, otherwise it's nil
 	accountData map[common.Hash][]byte                 // Keyed accounts for direct retrival (nil means deleted)
@@ -50,11 +49,10 @@ type diffLayer struct {
 
 // newDiffLayer creates a new diff on top of an existing snapshot, whether that's a low
 // level persistent database or a hierarchical diff already.
-func newDiffLayer(parent snapshot, number uint64, root common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
+func newDiffLayer(parent snapshot, root common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
 	// Create the new layer with some pre-allocated data segments
 	dl := &diffLayer{
 		parent:      parent,
-		number:      number,
 		root:        root,
 		accountData: accounts,
 		storageData: storage,
@@ -63,7 +61,6 @@ func newDiffLayer(parent snapshot, number uint64, root common.Hash, accounts map
 	for _, data := range accounts {
 		dl.memory += uint64(len(data))
 	}
-
 	// Fill the storage hashes and sort them for the iterator
 	dl.storageList = make(map[common.Hash][]common.Hash)
 
@@ -93,9 +90,18 @@ func newDiffLayer(parent snapshot, number uint64, root common.Hash, accounts map
 	return dl
 }
 
-// Info returns the block number and root hash for which this snapshot was made.
-func (dl *diffLayer) Info() (uint64, common.Hash) {
-	return dl.number, dl.root
+// Root returns the root hash for which this snapshot was made.
+func (dl *diffLayer) Root() common.Hash {
+	return dl.root
+}
+
+// Stale return whether this layer has become stale (was flattened across) or if
+// it's still live.
+func (dl *diffLayer) Stale() bool {
+	dl.lock.RLock()
+	defer dl.lock.RUnlock()
+
+	return dl.stale
 }
 
 // Account directly retrieves the account associated with a particular hash in
@@ -164,7 +170,7 @@ func (dl *diffLayer) Storage(accountHash, storageHash common.Hash) ([]byte, erro
 // Update creates a new layer on top of the existing snapshot diff tree with
 // the specified data items.
 func (dl *diffLayer) Update(blockRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
-	return newDiffLayer(dl, dl.number+1, blockRoot, accounts, storage)
+	return newDiffLayer(dl, blockRoot, accounts, storage)
 }
 
 // flatten pushes all data from this point downwards, flattening everything into
@@ -213,7 +219,6 @@ func (dl *diffLayer) flatten() snapshot {
 	// Return the combo parent
 	return &diffLayer{
 		parent:      parent.parent,
-		number:      dl.number,
 		root:        dl.root,
 		storageList: parent.storageList,
 		storageData: parent.storageData,

core/state/snapshot/difflayer_journal.go

@@ -43,18 +43,12 @@ type journalStorage struct {
 // diff and verifying that it can be linked to the requested parent.
 func loadDiffLayer(parent snapshot, r *rlp.Stream) (snapshot, error) {
 	// Read the next diff journal entry
-	var (
-		number uint64
-		root   common.Hash
-	)
-	if err := r.Decode(&number); err != nil {
+	var root common.Hash
+	if err := r.Decode(&root); err != nil {
 		// The first read may fail with EOF, marking the end of the journal
 		if err == io.EOF {
 			return parent, nil
 		}
-		return nil, fmt.Errorf("load diff number: %v", err)
-	}
-	if err := r.Decode(&root); err != nil {
 		return nil, fmt.Errorf("load diff root: %v", err)
 	}
 	var accounts []journalAccount
@@ -77,13 +71,7 @@ func loadDiffLayer(parent snapshot, r *rlp.Stream) (snapshot, error) {
 		}
 		storageData[entry.Hash] = slots
 	}
-	// Validate the block number to avoid state corruption
-	if parent, ok := parent.(*diffLayer); ok {
-		if number != parent.number+1 {
-			return nil, fmt.Errorf("snapshot chain broken: block #%d after #%d", number, parent.number)
-		}
-	}
-	return loadDiffLayer(newDiffLayer(parent, number, root, accountData, storageData), r)
+	return loadDiffLayer(newDiffLayer(parent, root, accountData, storageData), r)
 }
 
 // journal is the internal version of Journal that also returns the journal file
@@ -113,13 +101,8 @@ func (dl *diffLayer) journal() (io.WriteCloser, error) {
 		writer.Close()
 		return nil, ErrSnapshotStale
 	}
-	buf := bufio.NewWriter(writer)
 	// Everything below was journalled, persist this layer too
-	if err := rlp.Encode(buf, dl.number); err != nil {
-		buf.Flush()
-		writer.Close()
-		return nil, err
-	}
+	buf := bufio.NewWriter(writer)
 	if err := rlp.Encode(buf, dl.root); err != nil {
 		buf.Flush()
 		writer.Close()

core/state/snapshot/difflayer_test.go

@@ -61,11 +61,11 @@ func TestMergeBasics(t *testing.T) {
 		}
 	}
 	// Add some (identical) layers on top
-	parent := newDiffLayer(emptyLayer{}, 1, common.Hash{}, accounts, storage)
-	child := newDiffLayer(parent, 1, common.Hash{}, accounts, storage)
-	child = newDiffLayer(child, 1, common.Hash{}, accounts, storage)
-	child = newDiffLayer(child, 1, common.Hash{}, accounts, storage)
-	child = newDiffLayer(child, 1, common.Hash{}, accounts, storage)
+	parent := newDiffLayer(emptyLayer{}, common.Hash{}, accounts, storage)
+	child := newDiffLayer(parent, common.Hash{}, accounts, storage)
+	child = newDiffLayer(child, common.Hash{}, accounts, storage)
+	child = newDiffLayer(child, common.Hash{}, accounts, storage)
+	child = newDiffLayer(child, common.Hash{}, accounts, storage)
 	// And flatten
 	merged := (child.flatten()).(*diffLayer)
 
@@ -122,7 +122,7 @@ func TestMergeDelete(t *testing.T) {
 	}
 
 	// Add some flip-flopping layers on top
-	parent := newDiffLayer(emptyLayer{}, 1, common.Hash{}, flip(), storage)
+	parent := newDiffLayer(emptyLayer{}, common.Hash{}, flip(), storage)
 	child := parent.Update(common.Hash{}, flop(), storage)
 	child = child.Update(common.Hash{}, flip(), storage)
 	child = child.Update(common.Hash{}, flop(), storage)
@@ -139,10 +139,6 @@ func TestMergeDelete(t *testing.T) {
 	// And flatten
 	merged := (child.flatten()).(*diffLayer)
 
-	// check number
-	if got, exp := merged.number, child.number; got != exp {
-		t.Errorf("merged layer: wrong number - exp %d got %d", exp, got)
-	}
 	if data, _ := merged.Account(h1); data == nil {
 		t.Errorf("merged layer: expected %x to be non-nil", h1)
 	}
@@ -169,7 +165,7 @@ func TestInsertAndMerge(t *testing.T) {
 	{
 		var accounts = make(map[common.Hash][]byte)
 		var storage = make(map[common.Hash]map[common.Hash][]byte)
-		parent = newDiffLayer(emptyLayer{}, 1, common.Hash{}, accounts, storage)
+		parent = newDiffLayer(emptyLayer{}, common.Hash{}, accounts, storage)
 	}
 	{
 		var accounts = make(map[common.Hash][]byte)
@@ -178,7 +174,7 @@ func TestInsertAndMerge(t *testing.T) {
 		accstorage := make(map[common.Hash][]byte)
 		storage[acc] = accstorage
 		storage[acc][slot] = []byte{0x01}
-		child = newDiffLayer(parent, 2, common.Hash{}, accounts, storage)
+		child = newDiffLayer(parent, common.Hash{}, accounts, storage)
 	}
 	// And flatten
 	merged := (child.flatten()).(*diffLayer)
@@ -200,11 +196,12 @@ func (emptyLayer) Journal() error {
 	panic("implement me")
 }
 
-func (emptyLayer) Info() (uint64, common.Hash) {
-	return 0, common.Hash{}
+func (emptyLayer) Stale() bool {
+	panic("implement me")
 }
-func (emptyLayer) Number() uint64 {
-	return 0
+
+func (emptyLayer) Root() common.Hash {
+	return common.Hash{}
 }
 
 func (emptyLayer) Account(hash common.Hash) (*Account, error) {
@@ -227,8 +224,6 @@ func (emptyLayer) Storage(accountHash, storageHash common.Hash) ([]byte, error)
 // BenchmarkSearch-6   	  500000	      3723 ns/op (10k per layer, only top-level RLock()
 func BenchmarkSearch(b *testing.B) {
 	// First, we set up 128 diff layers, with 1K items each
-
-	blocknum := uint64(0)
 	fill := func(parent snapshot) *diffLayer {
 		accounts := make(map[common.Hash][]byte)
 		storage := make(map[common.Hash]map[common.Hash][]byte)
@@ -236,10 +231,8 @@ func BenchmarkSearch(b *testing.B) {
 		for i := 0; i < 10000; i++ {
 			accounts[randomHash()] = randomAccount()
 		}
-		blocknum++
-		return newDiffLayer(parent, blocknum, common.Hash{}, accounts, storage)
+		return newDiffLayer(parent, common.Hash{}, accounts, storage)
 	}
-
 	var layer snapshot
 	layer = emptyLayer{}
 	for i := 0; i < 128; i++ {
@@ -261,8 +254,6 @@ func BenchmarkSearch(b *testing.B) {
 // BenchmarkSearchSlot-6   	  100000	     14551 ns/op (when checking parent number using atomic)
 func BenchmarkSearchSlot(b *testing.B) {
 	// First, we set up 128 diff layers, with 1K items each
-
-	blocknum := uint64(0)
 	accountKey := common.Hash{}
 	storageKey := common.HexToHash("0x1337")
 	accountRLP := randomAccount()
@@ -278,16 +269,13 @@ func BenchmarkSearchSlot(b *testing.B) {
 			accStorage[randomHash()] = value
 			storage[accountKey] = accStorage
 		}
-		blocknum++
-		return newDiffLayer(parent, blocknum, common.Hash{}, accounts, storage)
+		return newDiffLayer(parent, common.Hash{}, accounts, storage)
 	}
-
 	var layer snapshot
 	layer = emptyLayer{}
 	for i := 0; i < 128; i++ {
 		layer = fill(layer)
 	}
-
 	b.ResetTimer()
 	for i := 0; i < b.N; i++ {
 		layer.Storage(accountKey, storageKey)
@@ -300,7 +288,7 @@ func BenchmarkSearchSlot(b *testing.B) {
 // Without sorting and tracking accountlist
 // BenchmarkFlatten-6   	     300	   5511511 ns/op
 func BenchmarkFlatten(b *testing.B) {
-	fill := func(parent snapshot, blocknum int) *diffLayer {
+	fill := func(parent snapshot) *diffLayer {
 		accounts := make(map[common.Hash][]byte)
 		storage := make(map[common.Hash]map[common.Hash][]byte)
 
@@ -317,7 +305,7 @@ func BenchmarkFlatten(b *testing.B) {
 			}
 			storage[accountKey] = accStorage
 		}
-		return newDiffLayer(parent, uint64(blocknum), common.Hash{}, accounts, storage)
+		return newDiffLayer(parent, common.Hash{}, accounts, storage)
 	}
 
 	b.ResetTimer()
@@ -327,7 +315,7 @@ func BenchmarkFlatten(b *testing.B) {
 		var layer snapshot
 		layer = emptyLayer{}
 		for i := 1; i < 128; i++ {
-			layer = fill(layer, i)
+			layer = fill(layer)
 		}
 		b.StartTimer()
 
@@ -336,7 +324,6 @@ func BenchmarkFlatten(b *testing.B) {
 			if !ok {
 				break
 			}
-
 			layer = dl.flatten()
 		}
 		b.StopTimer()
@@ -351,7 +338,7 @@ func BenchmarkFlatten(b *testing.B) {
 // BenchmarkJournal-6   	       1	1471373923 ns/ops
 // BenchmarkJournal-6   	       1	1208083335 ns/op // bufio writer
 func BenchmarkJournal(b *testing.B) {
-	fill := func(parent snapshot, blocknum int) *diffLayer {
+	fill := func(parent snapshot) *diffLayer {
 		accounts := make(map[common.Hash][]byte)
 		storage := make(map[common.Hash]map[common.Hash][]byte)
 
@@ -368,15 +355,14 @@ func BenchmarkJournal(b *testing.B) {
 			}
 			storage[accountKey] = accStorage
 		}
-		return newDiffLayer(parent, uint64(blocknum), common.Hash{}, accounts, storage)
+		return newDiffLayer(parent, common.Hash{}, accounts, storage)
 	}
-
 	var layer snapshot
 	layer = &diskLayer{
 		journal: path.Join(os.TempDir(), "difflayer_journal.tmp"),
 	}
 	for i := 1; i < 128; i++ {
-		layer = fill(layer, i)
+		layer = fill(layer)
 	}
 	b.ResetTimer()
 

core/state/snapshot/disklayer.go

@@ -32,16 +32,24 @@ type diskLayer struct {
 	db      ethdb.KeyValueStore // Key-value store containing the base snapshot
 	cache   *bigcache.BigCache  // Cache to avoid hitting the disk for direct access
 
-	number uint64      // Block number of the base snapshot
-	root   common.Hash // Root hash of the base snapshot
-	stale  bool        // Signals that the layer became stale (state progressed)
+	root  common.Hash // Root hash of the base snapshot
+	stale bool        // Signals that the layer became stale (state progressed)
 
 	lock sync.RWMutex
 }
 
-// Info returns the block number and root hash for which this snapshot was made.
-func (dl *diskLayer) Info() (uint64, common.Hash) {
-	return dl.number, dl.root
+// Root returns  root hash for which this snapshot was made.
+func (dl *diskLayer) Root() common.Hash {
+	return dl.root
+}
+
+// Stale return whether this layer has become stale (was flattened across) or if
+// it's still live.
+func (dl *diskLayer) Stale() bool {
+	dl.lock.RLock()
+	defer dl.lock.RUnlock()
+
+	return dl.stale
 }
 
 // Account directly retrieves the account associated with a particular hash in
@@ -123,7 +131,7 @@ func (dl *diskLayer) Storage(accountHash, storageHash common.Hash) ([]byte, erro
 // the specified data items. Note, the maps are retained by the method to avoid
 // copying everything.
 func (dl *diskLayer) Update(blockHash common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
-	return newDiffLayer(dl, dl.number+1, blockHash, accounts, storage)
+	return newDiffLayer(dl, blockHash, accounts, storage)
 }
 
 // Journal commits an entire diff hierarchy to disk into a single journal file.

core/state/snapshot/generate.go

@@ -85,7 +85,7 @@ func wipeSnapshot(db ethdb.KeyValueStore) error {
 	}
 	it.Release()
 
-	rawdb.DeleteSnapshotBlock(batch)
+	rawdb.DeleteSnapshotRoot(batch)
 	if err := batch.Write(); err != nil {
 		return err
 	}
@@ -107,7 +107,7 @@ func wipeSnapshot(db ethdb.KeyValueStore) error {
 }
 
 // generateSnapshot regenerates a brand new snapshot based on an existing state database and head block.
-func generateSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot common.Hash) (snapshot, error) {
+func generateSnapshot(db ethdb.KeyValueStore, journal string, root common.Hash) (snapshot, error) {
 	// Wipe any previously existing snapshot from the database
 	if err := wipeSnapshot(db); err != nil {
 		return nil, err
@@ -124,7 +124,7 @@ func generateSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64,
 	batch := db.NewBatch()
 	triedb := trie.NewDatabase(db)
 
-	accTrie, err := trie.NewSecure(headRoot, triedb)
+	accTrie, err := trie.NewSecure(root, triedb)
 	if err != nil {
 		return nil, err
 	}
@@ -186,7 +186,7 @@ func generateSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64,
 	fmt.Printf("Totals: %9s (%d accs, %d nodes) + %9s (%d slots, %d nodes)\n", accountSize.TerminalString(), accountCount, accIt.Nodes, storageSize.TerminalString(), storageCount, storageNodes)
 
 	// Update the snapshot block marker and write any remainder data
-	rawdb.WriteSnapshotBlock(batch, headNumber, headRoot)
+	rawdb.WriteSnapshotRoot(batch, root)
 	batch.Write()
 	batch.Reset()
 
@@ -207,7 +207,6 @@ func generateSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64,
 		journal: journal,
 		db:      db,
 		cache:   cache,
-		number:  headNumber,
-		root:    headRoot,
+		root:    root,
 	}, nil
 }

core/state/snapshot/generate_test.go

@@ -47,7 +47,7 @@ func TestWipe(t *testing.T) {
 			rawdb.WriteStorageSnapshot(db, account, randomHash(), randomHash().Bytes())
 		}
 	}
-	rawdb.WriteSnapshotBlock(db, 123, randomHash())
+	rawdb.WriteSnapshotRoot(db, randomHash())
 
 	// Add some random non-snapshot data too to make wiping harder
 	for i := 0; i < 65536; i++ {
@@ -76,8 +76,8 @@ func TestWipe(t *testing.T) {
 	if items != 128+128*1024 {
 		t.Fatalf("snapshot size mismatch: have %d, want %d", items, 128+128*1024)
 	}
-	if number, hash := rawdb.ReadSnapshotBlock(db); number != 123 || hash == (common.Hash{}) {
-		t.Errorf("snapshot block marker mismatch: have #%d [%#x], want #%d [<not-nil>]", number, hash, 123)
+	if hash := rawdb.ReadSnapshotRoot(db); hash == (common.Hash{}) {
+		t.Errorf("snapshot block marker mismatch: have %#x, want <not-nil>", hash)
 	}
 	// Wipe all snapshot entries from the database
 	if err := wipeSnapshot(db); err != nil {
@@ -93,8 +93,8 @@ func TestWipe(t *testing.T) {
 			t.Errorf("snapshot entry remained after wipe: %x", key)
 		}
 	}
-	if number, hash := rawdb.ReadSnapshotBlock(db); number != 0 || hash != (common.Hash{}) {
-		t.Errorf("snapshot block marker remained after wipe: #%d [%#x]", number, hash)
+	if hash := rawdb.ReadSnapshotRoot(db); hash != (common.Hash{}) {
+		t.Errorf("snapshot block marker remained after wipe: %#x", hash)
 	}
 	// Iterate over the database and ensure miscellaneous items are present
 	items = 0

core/state/snapshot/snapshot.go

@@ -43,12 +43,16 @@ var (
 	// layer had been invalidated due to the chain progressing forward far enough
 	// to not maintain the layer's original state.
 	ErrSnapshotStale = errors.New("snapshot stale")
+
+	// errSnapshotCycle is returned if a snapshot is attempted to be inserted
+	// that forms a cycle in the snapshot tree.
+	errSnapshotCycle = errors.New("snapshot cycle")
 )
 
 // Snapshot represents the functionality supported by a snapshot storage layer.
 type Snapshot interface {
-	// Info returns the block number and root hash for which this snapshot was made.
-	Info() (uint64, common.Hash)
+	// Root returns the root hash for which this snapshot was made.
+	Root() common.Hash
 
 	// Account directly retrieves the account associated with a particular hash in
 	// the snapshot slim data format.
@@ -77,6 +81,10 @@ type snapshot interface {
 	// This is meant to be used during shutdown to persist the snapshot without
 	// flattening everything down (bad for reorgs).
 	Journal() error
+
+	// Stale return whether this layer has become stale (was flattened across) or
+	// if it's still live.
+	Stale() bool
 }
 
 // SnapshotTree is an Ethereum state snapshot tree. It consists of one persistent
@@ -88,7 +96,7 @@ type snapshot interface {
 // The goal of a state snapshot is twofold: to allow direct access to account and
 // storage data to avoid expensive multi-level trie lookups; and to allow sorted,
 // cheap iteration of the account/storage tries for sync aid.
-type SnapshotTree struct {
+type Tree struct {
 	layers map[common.Hash]snapshot // Collection of all known layers // TODO(karalabe): split Clique overlaps
 	lock   sync.RWMutex
 }
@@ -99,22 +107,21 @@ type SnapshotTree struct {
 //
 // If the snapshot is missing or inconsistent, the entirety is deleted and will
 // be reconstructed from scratch based on the tries in the key-value store.
-func New(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot common.Hash) (*SnapshotTree, error) {
+func New(db ethdb.KeyValueStore, journal string, root common.Hash) (*Tree, error) {
 	// Attempt to load a previously persisted snapshot
-	head, err := loadSnapshot(db, journal, headNumber, headRoot)
+	head, err := loadSnapshot(db, journal, root)
 	if err != nil {
 		log.Warn("Failed to load snapshot, regenerating", "err", err)
-		if head, err = generateSnapshot(db, journal, headNumber, headRoot); err != nil {
+		if head, err = generateSnapshot(db, journal, root); err != nil {
 			return nil, err
 		}
 	}
 	// Existing snapshot loaded or one regenerated, seed all the layers
-	snap := &SnapshotTree{
+	snap := &Tree{
 		layers: make(map[common.Hash]snapshot),
 	}
 	for head != nil {
-		_, root := head.Info()
-		snap.layers[root] = head
+		snap.layers[head.Root()] = head
 
 		switch self := head.(type) {
 		case *diffLayer:
@@ -130,54 +137,57 @@ func New(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot com
 
 // Snapshot retrieves a snapshot belonging to the given block root, or nil if no
 // snapshot is maintained for that block.
-func (st *SnapshotTree) Snapshot(blockRoot common.Hash) Snapshot {
-	st.lock.RLock()
-	defer st.lock.RUnlock()
+func (t *Tree) Snapshot(blockRoot common.Hash) Snapshot {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
 
-	return st.layers[blockRoot]
+	return t.layers[blockRoot]
 }
 
 // Update adds a new snapshot into the tree, if that can be linked to an existing
 // old parent. It is disallowed to insert a disk layer (the origin of all).
-func (st *SnapshotTree) Update(blockRoot common.Hash, parentRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) error {
+func (t *Tree) Update(blockRoot common.Hash, parentRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) error {
+	// Reject noop updates to avoid self-loops in the snapshot tree. This is a
+	// special case that can only happen for Clique networks where empty blocks
+	// don't modify the state (0 block subsidy).
+	//
+	// Although we could silently ignore this internally, it should be the caller's
+	// responsibility to avoid even attempting to insert such a snapshot.
+	if blockRoot == parentRoot {
+		return errSnapshotCycle
+	}
 	// Generate a new snapshot on top of the parent
-	parent := st.Snapshot(parentRoot).(snapshot)
+	parent := t.Snapshot(parentRoot).(snapshot)
 	if parent == nil {
 		return fmt.Errorf("parent [%#x] snapshot missing", parentRoot)
 	}
 	snap := parent.Update(blockRoot, accounts, storage)
 
 	// Save the new snapshot for later
-	st.lock.Lock()
-	defer st.lock.Unlock()
+	t.lock.Lock()
+	defer t.lock.Unlock()
 
-	st.layers[snap.root] = snap
+	t.layers[snap.root] = snap
 	return nil
 }
 
 // Cap traverses downwards the snapshot tree from a head block hash until the
 // number of allowed layers are crossed. All layers beyond the permitted number
 // are flattened downwards.
-func (st *SnapshotTree) Cap(blockRoot common.Hash, layers int, memory uint64) error {
+func (t *Tree) Cap(root common.Hash, layers int, memory uint64) error {
 	// Retrieve the head snapshot to cap from
-	var snap snapshot
-	if s := st.Snapshot(blockRoot); s == nil {
-		return fmt.Errorf("snapshot [%#x] missing", blockRoot)
-	} else {
-		snap = s.(snapshot)
+	snap := t.Snapshot(root)
+	if snap == nil {
+		return fmt.Errorf("snapshot [%#x] missing", root)
 	}
 	diff, ok := snap.(*diffLayer)
 	if !ok {
-		return fmt.Errorf("snapshot [%#x] is disk layer", blockRoot)
+		return fmt.Errorf("snapshot [%#x] is disk layer", root)
 	}
 	// Run the internal capping and discard all stale layers
-	st.lock.Lock()
-	defer st.lock.Unlock()
+	t.lock.Lock()
+	defer t.lock.Unlock()
 
-	var (
-		diskNumber uint64
-		diffNumber uint64
-	)
 	// Flattening the bottom-most diff layer requires special casing since there's
 	// no child to rewire to the grandparent. In that case we can fake a temporary
 	// child for the capping and then remove it.
@@ -188,8 +198,9 @@ func (st *SnapshotTree) Cap(blockRoot common.Hash, layers int, memory uint64) er
 		base := diffToDisk(diff.flatten().(*diffLayer))
 		diff.lock.RUnlock()
 
-		st.layers[base.root] = base
-		diskNumber, diffNumber = base.number, base.number
+		// Replace the entire snapshot tree with the flat base
+		t.layers = map[common.Hash]snapshot{base.root: base}
+		return nil
 
 	case 1:
 		// If full flattening was requested, flatten the diffs but only merge if the
@@ -205,59 +216,74 @@ func (st *SnapshotTree) Cap(blockRoot common.Hash, layers int, memory uint64) er
 		}
 		diff.lock.RUnlock()
 
+		// If all diff layers were removed, replace the entire snapshot tree
 		if base != nil {
-			st.layers[base.root] = base
-			diskNumber, diffNumber = base.number, base.number
-		} else {
-			st.layers[bottom.root] = bottom
-			diskNumber, diffNumber = bottom.parent.(*diskLayer).number, bottom.number
+			t.layers = map[common.Hash]snapshot{base.root: base}
+			return nil
 		}
+		// Merge the new aggregated layer into the snapshot tree, clean stales below
+		t.layers[bottom.root] = bottom
 
 	default:
-		diskNumber, diffNumber = st.cap(diff, layers, memory)
+		// Many layers requested to be retained, cap normally
+		t.cap(diff, layers, memory)
+	}
+	// Remove any layer that is stale or links into a stale layer
+	children := make(map[common.Hash][]common.Hash)
+	for root, snap := range t.layers {
+		if diff, ok := snap.(*diffLayer); ok {
+			parent := diff.parent.Root()
+			children[parent] = append(children[parent], root)
+		}
 	}
-	for root, snap := range st.layers {
-		if number, _ := snap.Info(); number != diskNumber && number < diffNumber {
-			delete(st.layers, root)
+	var remove func(root common.Hash)
+	remove = func(root common.Hash) {
+		delete(t.layers, root)
+		for _, child := range children[root] {
+			remove(child)
+		}
+		delete(children, root)
+	}
+	for root, snap := range t.layers {
+		if snap.Stale() {
+			remove(root)
 		}
 	}
 	return nil
 }
 
 // cap traverses downwards the diff tree until the number of allowed layers are
-// crossed. All diffs beyond the permitted number are flattened downwards. If
-// the layer limit is reached, memory cap is also enforced (but not before). The
-// block numbers for the disk layer and first diff layer are returned for GC.
-func (st *SnapshotTree) cap(diff *diffLayer, layers int, memory uint64) (uint64, uint64) {
+// crossed. All diffs beyond the permitted number are flattened downwards. If the
+// layer limit is reached, memory cap is also enforced (but not before).
+func (t *Tree) cap(diff *diffLayer, layers int, memory uint64) {
 	// Dive until we run out of layers or reach the persistent database
 	for ; layers > 2; layers-- {
 		// If we still have diff layers below, continue down
 		if parent, ok := diff.parent.(*diffLayer); ok {
 			diff = parent
 		} else {
-			// Diff stack too shallow, return block numbers without modifications
-			return diff.parent.(*diskLayer).number, diff.number
+			// Diff stack too shallow, return without modifications
+			return
 		}
 	}
 	// We're out of layers, flatten anything below, stopping if it's the disk or if
 	// the memory limit is not yet exceeded.
 	switch parent := diff.parent.(type) {
 	case *diskLayer:
-		return parent.number, diff.number
+		return
 
 	case *diffLayer:
 		// Flatten the parent into the grandparent. The flattening internally obtains a
 		// write lock on grandparent.
 		flattened := parent.flatten().(*diffLayer)
-		st.layers[flattened.root] = flattened
+		t.layers[flattened.root] = flattened
 
 		diff.lock.Lock()
 		defer diff.lock.Unlock()
 
 		diff.parent = flattened
 		if flattened.memory < memory {
-			diskNumber, _ := flattened.parent.Info()
-			return diskNumber, flattened.number
+			return
 		}
 	default:
 		panic(fmt.Sprintf("unknown data layer: %T", parent))
@@ -269,10 +295,8 @@ func (st *SnapshotTree) cap(diff *diffLayer, layers int, memory uint64) (uint64,
 	base := diffToDisk(bottom)
 	bottom.lock.RUnlock()
 
-	st.layers[base.root] = base
+	t.layers[base.root] = base
 	diff.parent = base
-
-	return base.number, diff.number
 }
 
 // diffToDisk merges a bottom-most diff into the persistent disk layer underneath
@@ -284,7 +308,7 @@ func diffToDisk(bottom *diffLayer) *diskLayer {
 	)
 	// Start by temporarily deleting the current snapshot block marker. This
 	// ensures that in the case of a crash, the entire snapshot is invalidated.
-	rawdb.DeleteSnapshotBlock(batch)
+	rawdb.DeleteSnapshotRoot(batch)
 
 	// Mark the original base as stale as we're going to create a new wrapper
 	base.lock.Lock()
@@ -341,13 +365,12 @@ func diffToDisk(bottom *diffLayer) *diskLayer {
 		}
 	}
 	// Update the snapshot block marker and write any remainder data
-	rawdb.WriteSnapshotBlock(batch, bottom.number, bottom.root)
+	rawdb.WriteSnapshotRoot(batch, bottom.root)
 	if err := batch.Write(); err != nil {
 		log.Crit("Failed to write leftover snapshot", "err", err)
 	}
 	return &diskLayer{
 		root:    bottom.root,
-		number:  bottom.number,
 		cache:   base.cache,
 		db:      base.db,
 		journal: base.journal,
@@ -357,27 +380,25 @@ func diffToDisk(bottom *diffLayer) *diskLayer {
 // Journal commits an entire diff hierarchy to disk into a single journal file.
 // This is meant to be used during shutdown to persist the snapshot without
 // flattening everything down (bad for reorgs).
-func (st *SnapshotTree) Journal(blockRoot common.Hash) error {
-	// Retrieve the head snapshot to journal from
-	var snap snapshot
-	if s := st.Snapshot(blockRoot); s == nil {
+func (t *Tree) Journal(blockRoot common.Hash) error {
+	// Retrieve the head snapshot to journal from var snap snapshot
+	snap := t.Snapshot(blockRoot)
+	if snap == nil {
 		return fmt.Errorf("snapshot [%#x] missing", blockRoot)
-	} else {
-		snap = s.(snapshot)
 	}
 	// Run the journaling
-	st.lock.Lock()
-	defer st.lock.Unlock()
+	t.lock.Lock()
+	defer t.lock.Unlock()
 
-	return snap.Journal()
+	return snap.(snapshot).Journal()
 }
 
 // loadSnapshot loads a pre-existing state snapshot backed by a key-value store.
-func loadSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot common.Hash) (snapshot, error) {
+func loadSnapshot(db ethdb.KeyValueStore, journal string, root common.Hash) (snapshot, error) {
 	// Retrieve the block number and hash of the snapshot, failing if no snapshot
 	// is present in the database (or crashed mid-update).
-	number, root := rawdb.ReadSnapshotBlock(db)
-	if root == (common.Hash{}) {
+	baseRoot := rawdb.ReadSnapshotRoot(db)
+	if baseRoot == (common.Hash{}) {
 		return nil, errors.New("missing or corrupted snapshot")
 	}
 	cache, _ := bigcache.NewBigCache(bigcache.Config{ // TODO(karalabe): dedup
@@ -391,16 +412,14 @@ func loadSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64, hea
 		journal: journal,
 		db:      db,
 		cache:   cache,
-		number:  number,
-		root:    root,
+		root:    baseRoot,
 	}
 	// Load all the snapshot diffs from the journal, failing if their chain is broken
 	// or does not lead from the disk snapshot to the specified head.
 	if _, err := os.Stat(journal); os.IsNotExist(err) {
 		// Journal doesn't exist, don't worry if it's not supposed to
-		if number != headNumber || root != headRoot {
-			return nil, fmt.Errorf("snapshot journal missing, head doesn't match snapshot: #%d [%#x] vs. #%d [%#x]",
-				headNumber, headRoot, number, root)
+		if baseRoot != root {
+			return nil, fmt.Errorf("snapshot journal missing, head doesn't match snapshot: have %#x, want %#x", baseRoot, root)
 		}
 		return base, nil
 	}
@@ -414,10 +433,8 @@ func loadSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64, hea
 	}
 	// Entire snapshot journal loaded, sanity check the head and return
 	// Journal doesn't exist, don't worry if it's not supposed to
-	number, root = snapshot.Info()
-	if number != headNumber || root != headRoot {
-		return nil, fmt.Errorf("head doesn't match snapshot: #%d [%#x] vs. #%d [%#x]",
-			headNumber, headRoot, number, root)
+	if head := snapshot.Root(); head != root {
+		return nil, fmt.Errorf("head doesn't match snapshot: have %#x, want %#x", head, root)
 	}
 	return snapshot, nil
 }

core/state/snapshot/snapshot_test.go

@@ -37,7 +37,7 @@ func TestDiskLayerExternalInvalidationFullFlatten(t *testing.T) {
 		root:  common.HexToHash("0x01"),
 		cache: cache,
 	}
-	snaps := &SnapshotTree{
+	snaps := &Tree{
 		layers: map[common.Hash]snapshot{
 			base.root: base,
 		},
@@ -83,7 +83,7 @@ func TestDiskLayerExternalInvalidationPartialFlatten(t *testing.T) {
 		root:  common.HexToHash("0x01"),
 		cache: cache,
 	}
-	snaps := &SnapshotTree{
+	snaps := &Tree{
 		layers: map[common.Hash]snapshot{
 			base.root: base,
 		},
@@ -132,7 +132,7 @@ func TestDiffLayerExternalInvalidationFullFlatten(t *testing.T) {
 		root:  common.HexToHash("0x01"),
 		cache: cache,
 	}
-	snaps := &SnapshotTree{
+	snaps := &Tree{
 		layers: map[common.Hash]snapshot{
 			base.root: base,
 		},
@@ -181,7 +181,7 @@ func TestDiffLayerExternalInvalidationPartialFlatten(t *testing.T) {
 		root:  common.HexToHash("0x01"),
 		cache: cache,
 	}
-	snaps := &SnapshotTree{
+	snaps := &Tree{
 		layers: map[common.Hash]snapshot{
 			base.root: base,
 		},
@@ -213,7 +213,6 @@ func TestDiffLayerExternalInvalidationPartialFlatten(t *testing.T) {
 	if got := len(snaps.layers); got != exp {
 		t.Errorf("layers modified, got %d exp %d", got, exp)
 	}
-
 	// Flatten the diff layer into the bottom accumulator
 	if err := snaps.Cap(common.HexToHash("0x04"), 2, 1024*1024); err != nil {
 		t.Fatalf("failed to flatten diff layer into accumulator: %v", err)
@@ -247,7 +246,7 @@ func TestPostCapBasicDataAccess(t *testing.T) {
 		root:  common.HexToHash("0x01"),
 		cache: cache,
 	}
-	snaps := &SnapshotTree{
+	snaps := &Tree{
 		layers: map[common.Hash]snapshot{
 			base.root: base,
 		},

core/state/statedb.go

@@ -68,7 +68,7 @@ type StateDB struct {
 	db   Database
 	trie Trie
 
-	snaps        *snapshot.SnapshotTree
+	snaps        *snapshot.Tree
 	snap         snapshot.Snapshot
 	snapAccounts map[common.Hash][]byte
 	snapStorage  map[common.Hash]map[common.Hash][]byte
@@ -117,7 +117,7 @@ type StateDB struct {
 }
 
 // Create a new state from a given trie.
-func New(root common.Hash, db Database, snaps *snapshot.SnapshotTree) (*StateDB, error) {
+func New(root common.Hash, db Database, snaps *snapshot.Tree) (*StateDB, error) {
 	tr, err := db.OpenTrie(root)
 	if err != nil {
 		return nil, err
@@ -840,12 +840,14 @@ func (s *StateDB) Commit(deleteEmptyObjects bool) (common.Hash, error) {
 		if metrics.EnabledExpensive {
 			defer func(start time.Time) { s.SnapshotCommits += time.Since(start) }(time.Now())
 		}
-		_, parentRoot := s.snap.Info()
-		if err := s.snaps.Update(root, parentRoot, s.snapAccounts, s.snapStorage); err != nil {
-			log.Warn("Failed to update snapshot tree", "from", parentRoot, "to", root, "err", err)
-		}
-		if err := s.snaps.Cap(root, 16, 4*1024*1024); err != nil {
-			log.Warn("Failed to cap snapshot tree", "root", root, "layers", 16, "memory", 4*1024*1024, "err", err)
+		// Only update if there's a state transition (skip empty Clique blocks)
+		if parent := s.snap.Root(); parent != root {
+			if err := s.snaps.Update(root, parent, s.snapAccounts, s.snapStorage); err != nil {
+				log.Warn("Failed to update snapshot tree", "from", parent, "to", root, "err", err)
+			}
+			if err := s.snaps.Cap(root, 16, 4*1024*1024); err != nil {
+				log.Warn("Failed to cap snapshot tree", "root", root, "layers", 16, "memory", 4*1024*1024, "err", err)
+			}
 		}
 		s.snap, s.snapAccounts, s.snapStorage = nil, nil, nil
 	}