core, eth: split the db blocks into headers and bodies

core/chain_manager_test.go

@@ -388,7 +388,10 @@ func makeChainWithDiff(genesis *types.Block, d []int, seed byte) []*types.Block
 func chm(genesis *types.Block, db common.Database) *ChainManager {
 	var eventMux event.TypeMux
 	bc := &ChainManager{chainDb: db, genesisBlock: genesis, eventMux: &eventMux, pow: FakePow{}}
-	bc.cache, _ = lru.New(100)
+	bc.headerCache, _ = lru.New(100)
+	bc.bodyCache, _ = lru.New(100)
+	bc.bodyRLPCache, _ = lru.New(100)
+	bc.blockCache, _ = lru.New(100)
 	bc.futureBlocks, _ = lru.New(100)
 	bc.processor = bproc{}
 	bc.ResetWithGenesisBlock(genesis)

core/chain_util.go

@@ -19,7 +19,6 @@ package core
 import (
 	"bytes"
 	"math/big"
-	"time"
 
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/types"
@@ -30,9 +29,14 @@ import (
 )
 
 var (
-	blockHashPre  = []byte("block-hash-")
+	headKey = []byte("LastBlock")
+
+	headerHashPre = []byte("header-hash-")
+	bodyHashPre   = []byte("body-hash-")
 	blockNumPre   = []byte("block-num-")
 	ExpDiffPeriod = big.NewInt(100000)
+
+	blockHashPre = []byte("block-hash-") // [deprecated by eth/63]
 )
 
 // CalcDifficulty is the difficulty adjustment algorithm. It returns
@@ -112,68 +116,212 @@ func CalcGasLimit(parent *types.Block) *big.Int {
 	return gl
 }
 
-// GetBlockByHash returns the block corresponding to the hash or nil if not found
-func GetBlockByHash(db common.Database, hash common.Hash) *types.Block {
-	data, _ := db.Get(append(blockHashPre, hash[:]...))
+// storageBody is the block body encoding used for the database.
+type storageBody struct {
+	Transactions []*types.Transaction
+	Uncles       []*types.Header
+}
+
+// GetHashByNumber retrieves a hash assigned to a canonical block number.
+func GetHashByNumber(db common.Database, number uint64) common.Hash {
+	data, _ := db.Get(append(blockNumPre, big.NewInt(int64(number)).Bytes()...))
+	if len(data) == 0 {
+		return common.Hash{}
+	}
+	return common.BytesToHash(data)
+}
+
+// GetHeadHash retrieves the hash of the current canonical head block.
+func GetHeadHash(db common.Database) common.Hash {
+	data, _ := db.Get(headKey)
+	if len(data) == 0 {
+		return common.Hash{}
+	}
+	return common.BytesToHash(data)
+}
+
+// GetHeaderRLPByHash retrieves a block header in its raw RLP database encoding,
+// or nil if the header's not found.
+func GetHeaderRLPByHash(db common.Database, hash common.Hash) []byte {
+	data, _ := db.Get(append(headerHashPre, hash[:]...))
+	return data
+}
+
+// GetHeaderByHash retrieves the block header corresponding to the hash, nil if
+// none found.
+func GetHeaderByHash(db common.Database, hash common.Hash) *types.Header {
+	data := GetHeaderRLPByHash(db, hash)
 	if len(data) == 0 {
 		return nil
 	}
-	var block types.StorageBlock
-	if err := rlp.Decode(bytes.NewReader(data), &block); err != nil {
-		glog.V(logger.Error).Infof("invalid block RLP for hash %x: %v", hash, err)
+	header := new(types.Header)
+	if err := rlp.Decode(bytes.NewReader(data), header); err != nil {
+		glog.V(logger.Error).Infof("invalid block header RLP for hash %x: %v", hash, err)
 		return nil
 	}
-	return (*types.Block)(&block)
+	return header
 }
 
-// GetBlockByHash returns the canonical block by number or nil if not found
+// GetBodyRLPByHash retrieves the block body (transactions and uncles) in RLP
+// encoding, and the associated total difficulty.
+func GetBodyRLPByHash(db common.Database, hash common.Hash) ([]byte, *big.Int) {
+	combo, _ := db.Get(append(bodyHashPre, hash[:]...))
+	if len(combo) == 0 {
+		return nil, nil
+	}
+	buffer := bytes.NewBuffer(combo)
+
+	td := new(big.Int)
+	if err := rlp.Decode(buffer, td); err != nil {
+		glog.V(logger.Error).Infof("invalid block td RLP for hash %x: %v", hash, err)
+		return nil, nil
+	}
+	return buffer.Bytes(), td
+}
+
+// GetBodyByHash retrieves the block body (transactons, uncles, total difficulty)
+// corresponding to the hash, nils if none found.
+func GetBodyByHash(db common.Database, hash common.Hash) ([]*types.Transaction, []*types.Header, *big.Int) {
+	data, td := GetBodyRLPByHash(db, hash)
+	if len(data) == 0 || td == nil {
+		return nil, nil, nil
+	}
+	body := new(storageBody)
+	if err := rlp.Decode(bytes.NewReader(data), body); err != nil {
+		glog.V(logger.Error).Infof("invalid block body RLP for hash %x: %v", hash, err)
+		return nil, nil, nil
+	}
+	return body.Transactions, body.Uncles, td
+}
+
+// GetBlockByHash retrieves an entire block corresponding to the hash, assembling
+// it back from the stored header and body.
+func GetBlockByHash(db common.Database, hash common.Hash) *types.Block {
+	// Retrieve the block header and body contents
+	header := GetHeaderByHash(db, hash)
+	if header == nil {
+		return nil
+	}
+	transactions, uncles, td := GetBodyByHash(db, hash)
+	if td == nil {
+		return nil
+	}
+	// Reassemble the block and return
+	block := types.NewBlockWithHeader(header).WithBody(transactions, uncles)
+	block.Td = td
+
+	return block
+}
+
+// GetBlockByNumber returns the canonical block by number or nil if not found.
 func GetBlockByNumber(db common.Database, number uint64) *types.Block {
 	key, _ := db.Get(append(blockNumPre, big.NewInt(int64(number)).Bytes()...))
 	if len(key) == 0 {
 		return nil
 	}
-
 	return GetBlockByHash(db, common.BytesToHash(key))
 }
 
-// WriteCanonNumber writes the canonical hash for the given block
-func WriteCanonNumber(db common.Database, block *types.Block) error {
-	key := append(blockNumPre, block.Number().Bytes()...)
-	err := db.Put(key, block.Hash().Bytes())
-	if err != nil {
+// WriteCanonNumber stores the canonical hash for the given block number.
+func WriteCanonNumber(db common.Database, hash common.Hash, number uint64) error {
+	key := append(blockNumPre, big.NewInt(int64(number)).Bytes()...)
+	if err := db.Put(key, hash.Bytes()); err != nil {
+		glog.Fatalf("failed to store number to hash mapping into database: %v", err)
 		return err
 	}
 	return nil
 }
 
-// WriteHead force writes the current head
+// WriteHead updates the head block of the chain database.
 func WriteHead(db common.Database, block *types.Block) error {
-	err := WriteCanonNumber(db, block)
-	if err != nil {
+	if err := WriteCanonNumber(db, block.Hash(), block.NumberU64()); err != nil {
+		glog.Fatalf("failed to store canonical number into database: %v", err)
 		return err
 	}
-	err = db.Put([]byte("LastBlock"), block.Hash().Bytes())
-	if err != nil {
+	if err := db.Put(headKey, block.Hash().Bytes()); err != nil {
+		glog.Fatalf("failed to store last block into database: %v", err)
 		return err
 	}
 	return nil
 }
 
-// WriteBlock writes a block to the database
-func WriteBlock(db common.Database, block *types.Block) error {
-	tstart := time.Now()
+// WriteHeader serializes a block header into the database.
+func WriteHeader(db common.Database, header *types.Header) error {
+	data, err := rlp.EncodeToBytes(header)
+	if err != nil {
+		return err
+	}
+	key := append(headerHashPre, header.Hash().Bytes()...)
+	if err := db.Put(key, data); err != nil {
+		glog.Fatalf("failed to store header into database: %v", err)
+		return err
+	}
+	glog.V(logger.Debug).Infof("stored header #%v [%x…]", header.Number, header.Hash().Bytes()[:4])
+	return nil
+}
 
-	enc, _ := rlp.EncodeToBytes((*types.StorageBlock)(block))
-	key := append(blockHashPre, block.Hash().Bytes()...)
-	err := db.Put(key, enc)
+// WriteBody serializes the body of a block into the database.
+func WriteBody(db common.Database, block *types.Block) error {
+	body, err := rlp.EncodeToBytes(&storageBody{block.Transactions(), block.Uncles()})
 	if err != nil {
-		glog.Fatal("db write fail:", err)
 		return err
 	}
+	td, err := rlp.EncodeToBytes(block.Td)
+	if err != nil {
+		return err
+	}
+	key := append(bodyHashPre, block.Hash().Bytes()...)
+	if err := db.Put(key, append(td, body...)); err != nil {
+		glog.Fatalf("failed to store block body into database: %v", err)
+		return err
+	}
+	glog.V(logger.Debug).Infof("stored block body #%v [%x…]", block.Number, block.Hash().Bytes()[:4])
+	return nil
+}
 
-	if glog.V(logger.Debug) {
-		glog.Infof("wrote block #%v %s. Took %v\n", block.Number(), common.PP(block.Hash().Bytes()), time.Since(tstart))
+// WriteBlock serializes a block into the database, header and body separately.
+func WriteBlock(db common.Database, block *types.Block) error {
+	// Store the body first to retain database consistency
+	if err := WriteBody(db, block); err != nil {
+		return err
 	}
+	// Store the header too, signaling full block ownership
+	if err := WriteHeader(db, block.Header()); err != nil {
+		return err
+	}
+	return nil
+}
+
+// DeleteHeader removes all block header data associated with a hash.
+func DeleteHeader(db common.Database, hash common.Hash) {
+	db.Delete(append(headerHashPre, hash.Bytes()...))
+}
+
+// DeleteBody removes all block body data associated with a hash.
+func DeleteBody(db common.Database, hash common.Hash) {
+	db.Delete(append(bodyHashPre, hash.Bytes()...))
+}
+
+// DeleteBlock removes all block data associated with a hash.
+func DeleteBlock(db common.Database, hash common.Hash) {
+	DeleteHeader(db, hash)
+	DeleteBody(db, hash)
+}
 
+// [deprecated by eth/63]
+// GetBlockByHashOld returns the old combined block corresponding to the hash
+// or nil if not found. This method is only used by the upgrade mechanism to
+// access the old combined block representation. It will be dropped after the
+// network transitions to eth/63.
+func GetBlockByHashOld(db common.Database, hash common.Hash) *types.Block {
+	data, _ := db.Get(append(blockHashPre, hash[:]...))
+	if len(data) == 0 {
 		return nil
+	}
+	var block types.StorageBlock
+	if err := rlp.Decode(bytes.NewReader(data), &block); err != nil {
+		glog.V(logger.Error).Infof("invalid block RLP for hash %x: %v", hash, err)
+		return nil
+	}
+	return (*types.Block)(&block)
 }

core/genesis.go

@@ -86,7 +86,7 @@ func WriteGenesisBlock(chainDb common.Database, reader io.Reader) (*types.Block,
 
 	if block := GetBlockByHash(chainDb, block.Hash()); block != nil {
 		glog.V(logger.Info).Infoln("Genesis block already in chain. Writing canonical number")
-		err := WriteCanonNumber(chainDb, block)
+		err := WriteCanonNumber(chainDb, block.Hash(), block.NumberU64())
 		if err != nil {
 			return nil, err
 		}

core/types/block.go

@@ -135,6 +135,7 @@ type Block struct {
 	ReceivedAt time.Time
 }
 
+// [deprecated by eth/63]
 // StorageBlock defines the RLP encoding of a Block stored in the
 // state database. The StorageBlock encoding contains fields that
 // would otherwise need to be recomputed.
@@ -147,6 +148,7 @@ type extblock struct {
 	Uncles []*Header
 }
 
+// [deprecated by eth/63]
 // "storage" block encoding. used for database.
 type storageblock struct {
 	Header *Header
@@ -268,6 +270,7 @@ func (b *Block) EncodeRLP(w io.Writer) error {
 	})
 }
 
+// [deprecated by eth/63]
 func (b *StorageBlock) DecodeRLP(s *rlp.Stream) error {
 	var sb storageblock
 	if err := s.Decode(&sb); err != nil {
@@ -277,6 +280,7 @@ func (b *StorageBlock) DecodeRLP(s *rlp.Stream) error {
 	return nil
 }
 
+// [deprecated by eth/63]
 func (b *StorageBlock) EncodeRLP(w io.Writer) error {
 	return rlp.Encode(w, storageblock{
 		Header: b.header,

eth/backend.go

@@ -18,6 +18,7 @@
 package eth
 
 import (
+	"bytes"
 	"crypto/ecdsa"
 	"encoding/json"
 	"fmt"
@@ -269,11 +270,7 @@ func New(config *Config) (*Ethereum, error) {
 		newdb = func(path string) (common.Database, error) { return ethdb.NewLDBDatabase(path, config.DatabaseCache) }
 	}
 
-	// attempt to merge database together, upgrading from an old version
-	if err := mergeDatabases(config.DataDir, newdb); err != nil {
-		return nil, err
-	}
-
+	// Open the chain database and perform any upgrades needed
 	chainDb, err := newdb(filepath.Join(config.DataDir, "chaindata"))
 	if err != nil {
 		return nil, fmt.Errorf("blockchain db err: %v", err)
@@ -281,6 +278,10 @@ func New(config *Config) (*Ethereum, error) {
 	if db, ok := chainDb.(*ethdb.LDBDatabase); ok {
 		db.Meter("eth/db/chaindata/")
 	}
+	if err := upgradeChainDatabase(chainDb); err != nil {
+		return nil, err
+	}
+
 	dappDb, err := newdb(filepath.Join(config.DataDir, "dapp"))
 	if err != nil {
 		return nil, fmt.Errorf("dapp db err: %v", err)
@@ -721,74 +722,55 @@ func saveBlockchainVersion(db common.Database, bcVersion int) {
 	}
 }
 
-// mergeDatabases when required merge old database layout to one single database
-func mergeDatabases(datadir string, newdb func(path string) (common.Database, error)) error {
-	// Check if already upgraded
-	data := filepath.Join(datadir, "chaindata")
-	if _, err := os.Stat(data); !os.IsNotExist(err) {
+// upgradeChainDatabase ensures that the chain database stores block split into
+// separate header and body entries.
+func upgradeChainDatabase(db common.Database) error {
+	// Short circuit if the head block is stored already as separate header and body
+	data, err := db.Get([]byte("LastBlock"))
+	if err != nil {
 		return nil
 	}
-	// make sure it's not just a clean path
-	chainPath := filepath.Join(datadir, "blockchain")
-	if _, err := os.Stat(chainPath); os.IsNotExist(err) {
+	head := common.BytesToHash(data)
+
+	if block := core.GetBlockByHashOld(db, head); block == nil {
 		return nil
 	}
-	glog.Infoln("Database upgrade required. Upgrading...")
+	// At least some of the database is still the old format, upgrade (skip the head block!)
+	glog.V(logger.Info).Info("Old database detected, upgrading...")
 
-	database, err := newdb(data)
-	if err != nil {
-		return fmt.Errorf("creating data db err: %v", err)
+	if db, ok := db.(*ethdb.LDBDatabase); ok {
+		blockPrefix := []byte("block-hash-")
+		for it := db.NewIterator(); it.Next(); {
+			// Skip anything other than a combined block
+			if !bytes.HasPrefix(it.Key(), blockPrefix) {
+				continue
 			}
-	defer database.Close()
-
-	// Migrate blocks
-	chainDb, err := newdb(chainPath)
-	if err != nil {
-		return fmt.Errorf("state db err: %v", err)
+			// Skip the head block (merge last to signal upgrade completion)
+			if bytes.HasSuffix(it.Key(), head.Bytes()) {
+				continue
 			}
-	defer chainDb.Close()
+			// Load the block, split and serialize (order!)
+			block := core.GetBlockByHashOld(db, common.BytesToHash(bytes.TrimPrefix(it.Key(), blockPrefix)))
 
-	if chain, ok := chainDb.(*ethdb.LDBDatabase); ok {
-		glog.Infoln("Merging blockchain database...")
-		it := chain.NewIterator()
-		for it.Next() {
-			database.Put(it.Key(), it.Value())
-		}
-		it.Release()
+			if err := core.WriteBody(db, block); err != nil {
+				return err
 			}
-
-	// Migrate state
-	stateDb, err := newdb(filepath.Join(datadir, "state"))
-	if err != nil {
-		return fmt.Errorf("state db err: %v", err)
+			if err := core.WriteHeader(db, block.Header()); err != nil {
+				return err
 			}
-	defer stateDb.Close()
-
-	if state, ok := stateDb.(*ethdb.LDBDatabase); ok {
-		glog.Infoln("Merging state database...")
-		it := state.NewIterator()
-		for it.Next() {
-			database.Put(it.Key(), it.Value())
+			if err := db.Delete(it.Key()); err != nil {
+				return err
 			}
-		it.Release()
 		}
+		// Lastly, upgrade the head block, disabling the upgrade mechanism
+		current := core.GetBlockByHashOld(db, head)
 
-	// Migrate transaction / receipts
-	extraDb, err := newdb(filepath.Join(datadir, "extra"))
-	if err != nil {
-		return fmt.Errorf("state db err: %v", err)
+		if err := core.WriteBody(db, current); err != nil {
+			return err
 		}
-	defer extraDb.Close()
-
-	if extra, ok := extraDb.(*ethdb.LDBDatabase); ok {
-		glog.Infoln("Merging transaction database...")
-
-		it := extra.NewIterator()
-		for it.Next() {
-			database.Put(it.Key(), it.Value())
+		if err := core.WriteHeader(db, current.Header()); err != nil {
+			return err
 		}
-		it.Release()
 	}
-
 	return nil
 }

eth/handler.go

@@ -345,33 +345,33 @@ func (pm *ProtocolManager) handleMsg(p *peer) error {
 		if err := msg.Decode(&query); err != nil {
 			return errResp(ErrDecode, "%v: %v", msg, err)
 		}
-		// Gather blocks until the fetch or network limits is reached
+		// Gather headers until the fetch or network limits is reached
 		var (
 			bytes   common.StorageSize
 			headers []*types.Header
 			unknown bool
 		)
 		for !unknown && len(headers) < int(query.Amount) && bytes < softResponseLimit && len(headers) < downloader.MaxHeaderFetch {
-			// Retrieve the next block satisfying the query
-			var origin *types.Block
+			// Retrieve the next header satisfying the query
+			var origin *types.Header
 			if query.Origin.Hash != (common.Hash{}) {
-				origin = pm.chainman.GetBlock(query.Origin.Hash)
+				origin = pm.chainman.GetHeader(query.Origin.Hash)
 			} else {
-				origin = pm.chainman.GetBlockByNumber(query.Origin.Number)
+				origin = pm.chainman.GetHeaderByNumber(query.Origin.Number)
 			}
 			if origin == nil {
 				break
 			}
-			headers = append(headers, origin.Header())
-			bytes += origin.Size()
+			headers = append(headers, origin)
+			bytes += 500 // Approximate, should be good enough estimate
 
-			// Advance to the next block of the query
+			// Advance to the next header of the query
 			switch {
 			case query.Origin.Hash != (common.Hash{}) && query.Reverse:
 				// Hash based traversal towards the genesis block
 				for i := 0; i < int(query.Skip)+1; i++ {
-					if block := pm.chainman.GetBlock(query.Origin.Hash); block != nil {
-						query.Origin.Hash = block.ParentHash()
+					if header := pm.chainman.GetHeader(query.Origin.Hash); header != nil {
+						query.Origin.Hash = header.ParentHash
 					} else {
 						unknown = true
 						break
@@ -379,9 +379,9 @@ func (pm *ProtocolManager) handleMsg(p *peer) error {
 				}
 			case query.Origin.Hash != (common.Hash{}) && !query.Reverse:
 				// Hash based traversal towards the leaf block
-				if block := pm.chainman.GetBlockByNumber(origin.NumberU64() + query.Skip + 1); block != nil {
-					if pm.chainman.GetBlockHashesFromHash(block.Hash(), query.Skip+1)[query.Skip] == query.Origin.Hash {
-						query.Origin.Hash = block.Hash()
+				if header := pm.chainman.GetHeaderByNumber(origin.Number.Uint64() + query.Skip + 1); header != nil {
+					if pm.chainman.GetBlockHashesFromHash(header.Hash(), query.Skip+1)[query.Skip] == query.Origin.Hash {
+						query.Origin.Hash = header.Hash()
 					} else {
 						unknown = true
 					}
@@ -452,23 +452,24 @@ func (pm *ProtocolManager) handleMsg(p *peer) error {
 		// Gather blocks until the fetch or network limits is reached
 		var (
 			hash   common.Hash
-			bytes  common.StorageSize
-			bodies []*blockBody
+			bytes  int
+			bodies []*blockBodyRLP
 		)
 		for bytes < softResponseLimit && len(bodies) < downloader.MaxBlockFetch {
-			//Retrieve the hash of the next block
+			// Retrieve the hash of the next block
 			if err := msgStream.Decode(&hash); err == rlp.EOL {
 				break
 			} else if err != nil {
 				return errResp(ErrDecode, "msg %v: %v", msg, err)
 			}
-			// Retrieve the requested block, stopping if enough was found
-			if block := pm.chainman.GetBlock(hash); block != nil {
-				bodies = append(bodies, &blockBody{Transactions: block.Transactions(), Uncles: block.Uncles()})
-				bytes += block.Size()
+			// Retrieve the requested block body, stopping if enough was found
+			if data := pm.chainman.GetBodyRLP(hash); len(data) != 0 {
+				body := blockBodyRLP(data)
+				bodies = append(bodies, &body)
+				bytes += len(body)
 			}
 		}
-		return p.SendBlockBodies(bodies)
+		return p.SendBlockBodiesRLP(bodies)
 
 	case p.version >= eth63 && msg.Code == GetNodeDataMsg:
 		// Decode the retrieval message

eth/peer.go

@@ -184,6 +184,12 @@ func (p *peer) SendBlockBodies(bodies []*blockBody) error {
 	return p2p.Send(p.rw, BlockBodiesMsg, blockBodiesData(bodies))
 }
 
+// SendBlockBodiesRLP sends a batch of block contents to the remote peer from
+// an already RLP encoded format.
+func (p *peer) SendBlockBodiesRLP(bodies []*blockBodyRLP) error {
+	return p2p.Send(p.rw, BlockBodiesMsg, blockBodiesRLPData(bodies))
+}
+
 // SendNodeData sends a batch of arbitrary internal data, corresponding to the
 // hashes requested.
 func (p *peer) SendNodeData(data [][]byte) error {

eth/protocol.go

@@ -213,6 +213,22 @@ type blockBody struct {
 // blockBodiesData is the network packet for block content distribution.
 type blockBodiesData []*blockBody
 
+// blockBodyRLP represents the RLP encoded data content of a single block.
+type blockBodyRLP []byte
+
+// EncodeRLP is a specialized encoder for a block body to pass the already
+// encoded body RLPs from the database on, without double encoding.
+func (b *blockBodyRLP) EncodeRLP(w io.Writer) error {
+	if _, err := w.Write([]byte(*b)); err != nil {
+		return err
+	}
+	return nil
+}
+
+// blockBodiesRLPData is the network packet for block content distribution
+// based on original RLP formatting (i.e. skip the db-decode/proto-encode).
+type blockBodiesRLPData []*blockBodyRLP
+
 // nodeDataData is the network response packet for a node data retrieval.
 type nodeDataData []struct {
 	Value []byte