github.com/dotlike13/wemix30_go@v1.8.23/trie/database.go (about)

     1  // Copyright 2018 The go-ethereum Authors
     2  // This file is part of the go-ethereum library.
     3  //
     4  // The go-ethereum library is free software: you can redistribute it and/or modify
     5  // it under the terms of the GNU Lesser General Public License as published by
     6  // the Free Software Foundation, either version 3 of the License, or
     7  // (at your option) any later version.
     8  //
     9  // The go-ethereum library is distributed in the hope that it will be useful,
    10  // but WITHOUT ANY WARRANTY; without even the implied warranty of
    11  // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
    12  // GNU Lesser General Public License for more details.
    13  //
    14  // You should have received a copy of the GNU Lesser General Public License
    15  // along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
    16  
    17  package trie
    18  
    19  import (
    20  	"fmt"
    21  	"io"
    22  	"sync"
    23  	"time"
    24  
    25  	"github.com/allegro/bigcache"
    26  	"github.com/ethereum/go-ethereum/common"
    27  	"github.com/ethereum/go-ethereum/ethdb"
    28  	"github.com/ethereum/go-ethereum/log"
    29  	"github.com/ethereum/go-ethereum/metrics"
    30  	"github.com/ethereum/go-ethereum/rlp"
    31  )
    32  
    33  var (
    34  	memcacheCleanHitMeter   = metrics.NewRegisteredMeter("trie/memcache/clean/hit", nil)
    35  	memcacheCleanMissMeter  = metrics.NewRegisteredMeter("trie/memcache/clean/miss", nil)
    36  	memcacheCleanReadMeter  = metrics.NewRegisteredMeter("trie/memcache/clean/read", nil)
    37  	memcacheCleanWriteMeter = metrics.NewRegisteredMeter("trie/memcache/clean/write", nil)
    38  
    39  	memcacheFlushTimeTimer  = metrics.NewRegisteredResettingTimer("trie/memcache/flush/time", nil)
    40  	memcacheFlushNodesMeter = metrics.NewRegisteredMeter("trie/memcache/flush/nodes", nil)
    41  	memcacheFlushSizeMeter  = metrics.NewRegisteredMeter("trie/memcache/flush/size", nil)
    42  
    43  	memcacheGCTimeTimer  = metrics.NewRegisteredResettingTimer("trie/memcache/gc/time", nil)
    44  	memcacheGCNodesMeter = metrics.NewRegisteredMeter("trie/memcache/gc/nodes", nil)
    45  	memcacheGCSizeMeter  = metrics.NewRegisteredMeter("trie/memcache/gc/size", nil)
    46  
    47  	memcacheCommitTimeTimer  = metrics.NewRegisteredResettingTimer("trie/memcache/commit/time", nil)
    48  	memcacheCommitNodesMeter = metrics.NewRegisteredMeter("trie/memcache/commit/nodes", nil)
    49  	memcacheCommitSizeMeter  = metrics.NewRegisteredMeter("trie/memcache/commit/size", nil)
    50  )
    51  
    52  // secureKeyPrefix is the database key prefix used to store trie node preimages.
    53  var secureKeyPrefix = []byte("secure-key-")
    54  
    55  // secureKeyLength is the length of the above prefix + 32byte hash.
    56  const secureKeyLength = 11 + 32
    57  
    58  // DatabaseReader wraps the Get and Has method of a backing store for the trie.
    59  type DatabaseReader interface {
    60  	// Get retrieves the value associated with key from the database.
    61  	Get(key []byte) (value []byte, err error)
    62  
    63  	// Has retrieves whether a key is present in the database.
    64  	Has(key []byte) (bool, error)
    65  }
    66  
    67  // Database is an intermediate write layer between the trie data structures and
    68  // the disk database. The aim is to accumulate trie writes in-memory and only
    69  // periodically flush a couple tries to disk, garbage collecting the remainder.
    70  type Database struct {
    71  	diskdb ethdb.Database // Persistent storage for matured trie nodes
    72  
    73  	cleans  *bigcache.BigCache          // GC friendly memory cache of clean node RLPs
    74  	dirties map[common.Hash]*cachedNode // Data and references relationships of dirty nodes
    75  	oldest  common.Hash                 // Oldest tracked node, flush-list head
    76  	newest  common.Hash                 // Newest tracked node, flush-list tail
    77  
    78  	preimages map[common.Hash][]byte // Preimages of nodes from the secure trie
    79  	seckeybuf [secureKeyLength]byte  // Ephemeral buffer for calculating preimage keys
    80  
    81  	gctime  time.Duration      // Time spent on garbage collection since last commit
    82  	gcnodes uint64             // Nodes garbage collected since last commit
    83  	gcsize  common.StorageSize // Data storage garbage collected since last commit
    84  
    85  	flushtime  time.Duration      // Time spent on data flushing since last commit
    86  	flushnodes uint64             // Nodes flushed since last commit
    87  	flushsize  common.StorageSize // Data storage flushed since last commit
    88  
    89  	dirtiesSize   common.StorageSize // Storage size of the dirty node cache (exc. flushlist)
    90  	preimagesSize common.StorageSize // Storage size of the preimages cache
    91  
    92  	lock sync.RWMutex
    93  }
    94  
    95  // rawNode is a simple binary blob used to differentiate between collapsed trie
    96  // nodes and already encoded RLP binary blobs (while at the same time store them
    97  // in the same cache fields).
    98  type rawNode []byte
    99  
   100  func (n rawNode) canUnload(uint16, uint16) bool { panic("this should never end up in a live trie") }
   101  func (n rawNode) cache() (hashNode, bool)       { panic("this should never end up in a live trie") }
   102  func (n rawNode) fstring(ind string) string     { panic("this should never end up in a live trie") }
   103  
   104  // rawFullNode represents only the useful data content of a full node, with the
   105  // caches and flags stripped out to minimize its data storage. This type honors
   106  // the same RLP encoding as the original parent.
   107  type rawFullNode [17]node
   108  
   109  func (n rawFullNode) canUnload(uint16, uint16) bool { panic("this should never end up in a live trie") }
   110  func (n rawFullNode) cache() (hashNode, bool)       { panic("this should never end up in a live trie") }
   111  func (n rawFullNode) fstring(ind string) string     { panic("this should never end up in a live trie") }
   112  
   113  func (n rawFullNode) EncodeRLP(w io.Writer) error {
   114  	var nodes [17]node
   115  
   116  	for i, child := range n {
   117  		if child != nil {
   118  			nodes[i] = child
   119  		} else {
   120  			nodes[i] = nilValueNode
   121  		}
   122  	}
   123  	return rlp.Encode(w, nodes)
   124  }
   125  
   126  // rawShortNode represents only the useful data content of a short node, with the
   127  // caches and flags stripped out to minimize its data storage. This type honors
   128  // the same RLP encoding as the original parent.
   129  type rawShortNode struct {
   130  	Key []byte
   131  	Val node
   132  }
   133  
   134  func (n rawShortNode) canUnload(uint16, uint16) bool { panic("this should never end up in a live trie") }
   135  func (n rawShortNode) cache() (hashNode, bool)       { panic("this should never end up in a live trie") }
   136  func (n rawShortNode) fstring(ind string) string     { panic("this should never end up in a live trie") }
   137  
   138  // cachedNode is all the information we know about a single cached node in the
   139  // memory database write layer.
   140  type cachedNode struct {
   141  	node node   // Cached collapsed trie node, or raw rlp data
   142  	size uint16 // Byte size of the useful cached data
   143  
   144  	parents  uint32                 // Number of live nodes referencing this one
   145  	children map[common.Hash]uint16 // External children referenced by this node
   146  
   147  	flushPrev common.Hash // Previous node in the flush-list
   148  	flushNext common.Hash // Next node in the flush-list
   149  }
   150  
   151  // rlp returns the raw rlp encoded blob of the cached node, either directly from
   152  // the cache, or by regenerating it from the collapsed node.
   153  func (n *cachedNode) rlp() []byte {
   154  	if node, ok := n.node.(rawNode); ok {
   155  		return node
   156  	}
   157  	blob, err := rlp.EncodeToBytes(n.node)
   158  	if err != nil {
   159  		panic(err)
   160  	}
   161  	return blob
   162  }
   163  
   164  // obj returns the decoded and expanded trie node, either directly from the cache,
   165  // or by regenerating it from the rlp encoded blob.
   166  func (n *cachedNode) obj(hash common.Hash, cachegen uint16) node {
   167  	if node, ok := n.node.(rawNode); ok {
   168  		return mustDecodeNode(hash[:], node, cachegen)
   169  	}
   170  	return expandNode(hash[:], n.node, cachegen)
   171  }
   172  
   173  // childs returns all the tracked children of this node, both the implicit ones
   174  // from inside the node as well as the explicit ones from outside the node.
   175  func (n *cachedNode) childs() []common.Hash {
   176  	children := make([]common.Hash, 0, 16)
   177  	for child := range n.children {
   178  		children = append(children, child)
   179  	}
   180  	if _, ok := n.node.(rawNode); !ok {
   181  		gatherChildren(n.node, &children)
   182  	}
   183  	return children
   184  }
   185  
   186  // gatherChildren traverses the node hierarchy of a collapsed storage node and
   187  // retrieves all the hashnode children.
   188  func gatherChildren(n node, children *[]common.Hash) {
   189  	switch n := n.(type) {
   190  	case *rawShortNode:
   191  		gatherChildren(n.Val, children)
   192  
   193  	case rawFullNode:
   194  		for i := 0; i < 16; i++ {
   195  			gatherChildren(n[i], children)
   196  		}
   197  	case hashNode:
   198  		*children = append(*children, common.BytesToHash(n))
   199  
   200  	case valueNode, nil:
   201  
   202  	default:
   203  		panic(fmt.Sprintf("unknown node type: %T", n))
   204  	}
   205  }
   206  
   207  // simplifyNode traverses the hierarchy of an expanded memory node and discards
   208  // all the internal caches, returning a node that only contains the raw data.
   209  func simplifyNode(n node) node {
   210  	switch n := n.(type) {
   211  	case *shortNode:
   212  		// Short nodes discard the flags and cascade
   213  		return &rawShortNode{Key: n.Key, Val: simplifyNode(n.Val)}
   214  
   215  	case *fullNode:
   216  		// Full nodes discard the flags and cascade
   217  		node := rawFullNode(n.Children)
   218  		for i := 0; i < len(node); i++ {
   219  			if node[i] != nil {
   220  				node[i] = simplifyNode(node[i])
   221  			}
   222  		}
   223  		return node
   224  
   225  	case valueNode, hashNode, rawNode:
   226  		return n
   227  
   228  	default:
   229  		panic(fmt.Sprintf("unknown node type: %T", n))
   230  	}
   231  }
   232  
   233  // expandNode traverses the node hierarchy of a collapsed storage node and converts
   234  // all fields and keys into expanded memory form.
   235  func expandNode(hash hashNode, n node, cachegen uint16) node {
   236  	switch n := n.(type) {
   237  	case *rawShortNode:
   238  		// Short nodes need key and child expansion
   239  		return &shortNode{
   240  			Key: compactToHex(n.Key),
   241  			Val: expandNode(nil, n.Val, cachegen),
   242  			flags: nodeFlag{
   243  				hash: hash,
   244  				gen:  cachegen,
   245  			},
   246  		}
   247  
   248  	case rawFullNode:
   249  		// Full nodes need child expansion
   250  		node := &fullNode{
   251  			flags: nodeFlag{
   252  				hash: hash,
   253  				gen:  cachegen,
   254  			},
   255  		}
   256  		for i := 0; i < len(node.Children); i++ {
   257  			if n[i] != nil {
   258  				node.Children[i] = expandNode(nil, n[i], cachegen)
   259  			}
   260  		}
   261  		return node
   262  
   263  	case valueNode, hashNode:
   264  		return n
   265  
   266  	default:
   267  		panic(fmt.Sprintf("unknown node type: %T", n))
   268  	}
   269  }
   270  
   271  // NewDatabase creates a new trie database to store ephemeral trie content before
   272  // its written out to disk or garbage collected. No read cache is created, so all
   273  // data retrievals will hit the underlying disk database.
   274  func NewDatabase(diskdb ethdb.Database) *Database {
   275  	return NewDatabaseWithCache(diskdb, 0)
   276  }
   277  
   278  // NewDatabaseWithCache creates a new trie database to store ephemeral trie content
   279  // before its written out to disk or garbage collected. It also acts as a read cache
   280  // for nodes loaded from disk.
   281  func NewDatabaseWithCache(diskdb ethdb.Database, cache int) *Database {
   282  	var cleans *bigcache.BigCache
   283  	if cache > 0 {
   284  		cleans, _ = bigcache.NewBigCache(bigcache.Config{
   285  			Shards:             1024,
   286  			LifeWindow:         time.Hour,
   287  			MaxEntriesInWindow: cache * 1024,
   288  			MaxEntrySize:       512,
   289  			HardMaxCacheSize:   cache,
   290  		})
   291  	}
   292  	return &Database{
   293  		diskdb:    diskdb,
   294  		cleans:    cleans,
   295  		dirties:   map[common.Hash]*cachedNode{{}: {}},
   296  		preimages: make(map[common.Hash][]byte),
   297  	}
   298  }
   299  
   300  // DiskDB retrieves the persistent storage backing the trie database.
   301  func (db *Database) DiskDB() DatabaseReader {
   302  	return db.diskdb
   303  }
   304  
   305  // InsertBlob writes a new reference tracked blob to the memory database if it's
   306  // yet unknown. This method should only be used for non-trie nodes that require
   307  // reference counting, since trie nodes are garbage collected directly through
   308  // their embedded children.
   309  func (db *Database) InsertBlob(hash common.Hash, blob []byte) {
   310  	db.lock.Lock()
   311  	defer db.lock.Unlock()
   312  
   313  	db.insert(hash, blob, rawNode(blob))
   314  }
   315  
   316  // insert inserts a collapsed trie node into the memory database. This method is
   317  // a more generic version of InsertBlob, supporting both raw blob insertions as
   318  // well ex trie node insertions. The blob must always be specified to allow proper
   319  // size tracking.
   320  func (db *Database) insert(hash common.Hash, blob []byte, node node) {
   321  	// If the node's already cached, skip
   322  	if _, ok := db.dirties[hash]; ok {
   323  		return
   324  	}
   325  	// Create the cached entry for this node
   326  	entry := &cachedNode{
   327  		node:      simplifyNode(node),
   328  		size:      uint16(len(blob)),
   329  		flushPrev: db.newest,
   330  	}
   331  	for _, child := range entry.childs() {
   332  		if c := db.dirties[child]; c != nil {
   333  			c.parents++
   334  		}
   335  	}
   336  	db.dirties[hash] = entry
   337  
   338  	// Update the flush-list endpoints
   339  	if db.oldest == (common.Hash{}) {
   340  		db.oldest, db.newest = hash, hash
   341  	} else {
   342  		db.dirties[db.newest].flushNext, db.newest = hash, hash
   343  	}
   344  	db.dirtiesSize += common.StorageSize(common.HashLength + entry.size)
   345  }
   346  
   347  // insertPreimage writes a new trie node pre-image to the memory database if it's
   348  // yet unknown. The method will make a copy of the slice.
   349  //
   350  // Note, this method assumes that the database's lock is held!
   351  func (db *Database) insertPreimage(hash common.Hash, preimage []byte) {
   352  	if _, ok := db.preimages[hash]; ok {
   353  		return
   354  	}
   355  	db.preimages[hash] = common.CopyBytes(preimage)
   356  	db.preimagesSize += common.StorageSize(common.HashLength + len(preimage))
   357  }
   358  
   359  // node retrieves a cached trie node from memory, or returns nil if none can be
   360  // found in the memory cache.
   361  func (db *Database) node(hash common.Hash, cachegen uint16) node {
   362  	// Retrieve the node from the clean cache if available
   363  	if db.cleans != nil {
   364  		if enc, err := db.cleans.Get(string(hash[:])); err == nil && enc != nil {
   365  			memcacheCleanHitMeter.Mark(1)
   366  			memcacheCleanReadMeter.Mark(int64(len(enc)))
   367  			return mustDecodeNode(hash[:], enc, cachegen)
   368  		}
   369  	}
   370  	// Retrieve the node from the dirty cache if available
   371  	db.lock.RLock()
   372  	dirty := db.dirties[hash]
   373  	db.lock.RUnlock()
   374  
   375  	if dirty != nil {
   376  		return dirty.obj(hash, cachegen)
   377  	}
   378  	// Content unavailable in memory, attempt to retrieve from disk
   379  	enc, err := db.diskdb.Get(hash[:])
   380  	if err != nil || enc == nil {
   381  		return nil
   382  	}
   383  	if db.cleans != nil {
   384  		db.cleans.Set(string(hash[:]), enc)
   385  		memcacheCleanMissMeter.Mark(1)
   386  		memcacheCleanWriteMeter.Mark(int64(len(enc)))
   387  	}
   388  	return mustDecodeNode(hash[:], enc, cachegen)
   389  }
   390  
   391  // Node retrieves an encoded cached trie node from memory. If it cannot be found
   392  // cached, the method queries the persistent database for the content.
   393  func (db *Database) Node(hash common.Hash) ([]byte, error) {
   394  	// Retrieve the node from the clean cache if available
   395  	if db.cleans != nil {
   396  		if enc, err := db.cleans.Get(string(hash[:])); err == nil && enc != nil {
   397  			memcacheCleanHitMeter.Mark(1)
   398  			memcacheCleanReadMeter.Mark(int64(len(enc)))
   399  			return enc, nil
   400  		}
   401  	}
   402  	// Retrieve the node from the dirty cache if available
   403  	db.lock.RLock()
   404  	dirty := db.dirties[hash]
   405  	db.lock.RUnlock()
   406  
   407  	if dirty != nil {
   408  		return dirty.rlp(), nil
   409  	}
   410  	// Content unavailable in memory, attempt to retrieve from disk
   411  	enc, err := db.diskdb.Get(hash[:])
   412  	if err == nil && enc != nil {
   413  		if db.cleans != nil {
   414  			db.cleans.Set(string(hash[:]), enc)
   415  			memcacheCleanMissMeter.Mark(1)
   416  			memcacheCleanWriteMeter.Mark(int64(len(enc)))
   417  		}
   418  	}
   419  	return enc, err
   420  }
   421  
   422  // preimage retrieves a cached trie node pre-image from memory. If it cannot be
   423  // found cached, the method queries the persistent database for the content.
   424  func (db *Database) preimage(hash common.Hash) ([]byte, error) {
   425  	// Retrieve the node from cache if available
   426  	db.lock.RLock()
   427  	preimage := db.preimages[hash]
   428  	db.lock.RUnlock()
   429  
   430  	if preimage != nil {
   431  		return preimage, nil
   432  	}
   433  	// Content unavailable in memory, attempt to retrieve from disk
   434  	return db.diskdb.Get(db.secureKey(hash[:]))
   435  }
   436  
   437  // secureKey returns the database key for the preimage of key, as an ephemeral
   438  // buffer. The caller must not hold onto the return value because it will become
   439  // invalid on the next call.
   440  func (db *Database) secureKey(key []byte) []byte {
   441  	buf := append(db.seckeybuf[:0], secureKeyPrefix...)
   442  	buf = append(buf, key...)
   443  	return buf
   444  }
   445  
   446  // Nodes retrieves the hashes of all the nodes cached within the memory database.
   447  // This method is extremely expensive and should only be used to validate internal
   448  // states in test code.
   449  func (db *Database) Nodes() []common.Hash {
   450  	db.lock.RLock()
   451  	defer db.lock.RUnlock()
   452  
   453  	var hashes = make([]common.Hash, 0, len(db.dirties))
   454  	for hash := range db.dirties {
   455  		if hash != (common.Hash{}) { // Special case for "root" references/nodes
   456  			hashes = append(hashes, hash)
   457  		}
   458  	}
   459  	return hashes
   460  }
   461  
   462  // Reference adds a new reference from a parent node to a child node.
   463  func (db *Database) Reference(child common.Hash, parent common.Hash) {
   464  	db.lock.RLock()
   465  	defer db.lock.RUnlock()
   466  
   467  	db.reference(child, parent)
   468  }
   469  
   470  // reference is the private locked version of Reference.
   471  func (db *Database) reference(child common.Hash, parent common.Hash) {
   472  	// If the node does not exist, it's a node pulled from disk, skip
   473  	node, ok := db.dirties[child]
   474  	if !ok {
   475  		return
   476  	}
   477  	// If the reference already exists, only duplicate for roots
   478  	if db.dirties[parent].children == nil {
   479  		db.dirties[parent].children = make(map[common.Hash]uint16)
   480  	} else if _, ok = db.dirties[parent].children[child]; ok && parent != (common.Hash{}) {
   481  		return
   482  	}
   483  	node.parents++
   484  	db.dirties[parent].children[child]++
   485  }
   486  
   487  // Dereference removes an existing reference from a root node.
   488  func (db *Database) Dereference(root common.Hash) {
   489  	// Sanity check to ensure that the meta-root is not removed
   490  	if root == (common.Hash{}) {
   491  		log.Error("Attempted to dereference the trie cache meta root")
   492  		return
   493  	}
   494  	db.lock.Lock()
   495  	defer db.lock.Unlock()
   496  
   497  	nodes, storage, start := len(db.dirties), db.dirtiesSize, time.Now()
   498  	db.dereference(root, common.Hash{})
   499  
   500  	db.gcnodes += uint64(nodes - len(db.dirties))
   501  	db.gcsize += storage - db.dirtiesSize
   502  	db.gctime += time.Since(start)
   503  
   504  	memcacheGCTimeTimer.Update(time.Since(start))
   505  	memcacheGCSizeMeter.Mark(int64(storage - db.dirtiesSize))
   506  	memcacheGCNodesMeter.Mark(int64(nodes - len(db.dirties)))
   507  
   508  	log.Debug("Dereferenced trie from memory database", "nodes", nodes-len(db.dirties), "size", storage-db.dirtiesSize, "time", time.Since(start),
   509  		"gcnodes", db.gcnodes, "gcsize", db.gcsize, "gctime", db.gctime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
   510  }
   511  
   512  // dereference is the private locked version of Dereference.
   513  func (db *Database) dereference(child common.Hash, parent common.Hash) {
   514  	// Dereference the parent-child
   515  	node := db.dirties[parent]
   516  
   517  	if node.children != nil && node.children[child] > 0 {
   518  		node.children[child]--
   519  		if node.children[child] == 0 {
   520  			delete(node.children, child)
   521  		}
   522  	}
   523  	// If the child does not exist, it's a previously committed node.
   524  	node, ok := db.dirties[child]
   525  	if !ok {
   526  		return
   527  	}
   528  	// If there are no more references to the child, delete it and cascade
   529  	if node.parents > 0 {
   530  		// This is a special cornercase where a node loaded from disk (i.e. not in the
   531  		// memcache any more) gets reinjected as a new node (short node split into full,
   532  		// then reverted into short), causing a cached node to have no parents. That is
   533  		// no problem in itself, but don't make maxint parents out of it.
   534  		node.parents--
   535  	}
   536  	if node.parents == 0 {
   537  		// Remove the node from the flush-list
   538  		switch child {
   539  		case db.oldest:
   540  			db.oldest = node.flushNext
   541  			db.dirties[node.flushNext].flushPrev = common.Hash{}
   542  		case db.newest:
   543  			db.newest = node.flushPrev
   544  			db.dirties[node.flushPrev].flushNext = common.Hash{}
   545  		default:
   546  			db.dirties[node.flushPrev].flushNext = node.flushNext
   547  			db.dirties[node.flushNext].flushPrev = node.flushPrev
   548  		}
   549  		// Dereference all children and delete the node
   550  		for _, hash := range node.childs() {
   551  			db.dereference(hash, child)
   552  		}
   553  		delete(db.dirties, child)
   554  		db.dirtiesSize -= common.StorageSize(common.HashLength + int(node.size))
   555  	}
   556  }
   557  
   558  // Cap iteratively flushes old but still referenced trie nodes until the total
   559  // memory usage goes below the given threshold.
   560  func (db *Database) Cap(limit common.StorageSize) error {
   561  	// Create a database batch to flush persistent data out. It is important that
   562  	// outside code doesn't see an inconsistent state (referenced data removed from
   563  	// memory cache during commit but not yet in persistent storage). This is ensured
   564  	// by only uncaching existing data when the database write finalizes.
   565  	db.lock.RLock()
   566  
   567  	nodes, storage, start := len(db.dirties), db.dirtiesSize, time.Now()
   568  	batch := db.diskdb.NewBatch()
   569  
   570  	// db.dirtiesSize only contains the useful data in the cache, but when reporting
   571  	// the total memory consumption, the maintenance metadata is also needed to be
   572  	// counted. For every useful node, we track 2 extra hashes as the flushlist.
   573  	size := db.dirtiesSize + common.StorageSize((len(db.dirties)-1)*2*common.HashLength)
   574  
   575  	// If the preimage cache got large enough, push to disk. If it's still small
   576  	// leave for later to deduplicate writes.
   577  	flushPreimages := db.preimagesSize > 4*1024*1024
   578  	if flushPreimages {
   579  		for hash, preimage := range db.preimages {
   580  			if err := batch.Put(db.secureKey(hash[:]), preimage); err != nil {
   581  				log.Error("Failed to commit preimage from trie database", "err", err)
   582  				db.lock.RUnlock()
   583  				return err
   584  			}
   585  			if batch.ValueSize() > ethdb.IdealBatchSize {
   586  				if err := batch.Write(); err != nil {
   587  					db.lock.RUnlock()
   588  					return err
   589  				}
   590  				batch.Reset()
   591  			}
   592  		}
   593  	}
   594  	// Keep committing nodes from the flush-list until we're below allowance
   595  	oldest := db.oldest
   596  	for size > limit && oldest != (common.Hash{}) {
   597  		// Fetch the oldest referenced node and push into the batch
   598  		node := db.dirties[oldest]
   599  		if err := batch.Put(oldest[:], node.rlp()); err != nil {
   600  			db.lock.RUnlock()
   601  			return err
   602  		}
   603  		// If we exceeded the ideal batch size, commit and reset
   604  		if batch.ValueSize() >= ethdb.IdealBatchSize {
   605  			if err := batch.Write(); err != nil {
   606  				log.Error("Failed to write flush list to disk", "err", err)
   607  				db.lock.RUnlock()
   608  				return err
   609  			}
   610  			batch.Reset()
   611  		}
   612  		// Iterate to the next flush item, or abort if the size cap was achieved. Size
   613  		// is the total size, including both the useful cached data (hash -> blob), as
   614  		// well as the flushlist metadata (2*hash). When flushing items from the cache,
   615  		// we need to reduce both.
   616  		size -= common.StorageSize(3*common.HashLength + int(node.size))
   617  		oldest = node.flushNext
   618  	}
   619  	// Flush out any remainder data from the last batch
   620  	if err := batch.Write(); err != nil {
   621  		log.Error("Failed to write flush list to disk", "err", err)
   622  		db.lock.RUnlock()
   623  		return err
   624  	}
   625  	db.lock.RUnlock()
   626  
   627  	// Write successful, clear out the flushed data
   628  	db.lock.Lock()
   629  	defer db.lock.Unlock()
   630  
   631  	if flushPreimages {
   632  		db.preimages = make(map[common.Hash][]byte)
   633  		db.preimagesSize = 0
   634  	}
   635  	for db.oldest != oldest {
   636  		node := db.dirties[db.oldest]
   637  		delete(db.dirties, db.oldest)
   638  		db.oldest = node.flushNext
   639  
   640  		db.dirtiesSize -= common.StorageSize(common.HashLength + int(node.size))
   641  	}
   642  	if db.oldest != (common.Hash{}) {
   643  		db.dirties[db.oldest].flushPrev = common.Hash{}
   644  	}
   645  	db.flushnodes += uint64(nodes - len(db.dirties))
   646  	db.flushsize += storage - db.dirtiesSize
   647  	db.flushtime += time.Since(start)
   648  
   649  	memcacheFlushTimeTimer.Update(time.Since(start))
   650  	memcacheFlushSizeMeter.Mark(int64(storage - db.dirtiesSize))
   651  	memcacheFlushNodesMeter.Mark(int64(nodes - len(db.dirties)))
   652  
   653  	log.Debug("Persisted nodes from memory database", "nodes", nodes-len(db.dirties), "size", storage-db.dirtiesSize, "time", time.Since(start),
   654  		"flushnodes", db.flushnodes, "flushsize", db.flushsize, "flushtime", db.flushtime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
   655  
   656  	return nil
   657  }
   658  
   659  // Commit iterates over all the children of a particular node, writes them out
   660  // to disk, forcefully tearing down all references in both directions.
   661  //
   662  // As a side effect, all pre-images accumulated up to this point are also written.
   663  func (db *Database) Commit(node common.Hash, report bool) error {
   664  	// Create a database batch to flush persistent data out. It is important that
   665  	// outside code doesn't see an inconsistent state (referenced data removed from
   666  	// memory cache during commit but not yet in persistent storage). This is ensured
   667  	// by only uncaching existing data when the database write finalizes.
   668  	db.lock.RLock()
   669  
   670  	start := time.Now()
   671  	batch := db.diskdb.NewBatch()
   672  
   673  	// Move all of the accumulated preimages into a write batch
   674  	for hash, preimage := range db.preimages {
   675  		if err := batch.Put(db.secureKey(hash[:]), preimage); err != nil {
   676  			log.Error("Failed to commit preimage from trie database", "err", err)
   677  			db.lock.RUnlock()
   678  			return err
   679  		}
   680  		if batch.ValueSize() > ethdb.IdealBatchSize {
   681  			if err := batch.Write(); err != nil {
   682  				return err
   683  			}
   684  			batch.Reset()
   685  		}
   686  	}
   687  	// Move the trie itself into the batch, flushing if enough data is accumulated
   688  	nodes, storage := len(db.dirties), db.dirtiesSize
   689  	if err := db.commit(node, batch); err != nil {
   690  		log.Error("Failed to commit trie from trie database", "err", err)
   691  		db.lock.RUnlock()
   692  		return err
   693  	}
   694  	// Write batch ready, unlock for readers during persistence
   695  	if err := batch.Write(); err != nil {
   696  		log.Error("Failed to write trie to disk", "err", err)
   697  		db.lock.RUnlock()
   698  		return err
   699  	}
   700  	db.lock.RUnlock()
   701  
   702  	// Write successful, clear out the flushed data
   703  	db.lock.Lock()
   704  	defer db.lock.Unlock()
   705  
   706  	db.preimages = make(map[common.Hash][]byte)
   707  	db.preimagesSize = 0
   708  
   709  	db.uncache(node)
   710  
   711  	memcacheCommitTimeTimer.Update(time.Since(start))
   712  	memcacheCommitSizeMeter.Mark(int64(storage - db.dirtiesSize))
   713  	memcacheCommitNodesMeter.Mark(int64(nodes - len(db.dirties)))
   714  
   715  	logger := log.Info
   716  	if !report {
   717  		logger = log.Debug
   718  	}
   719  	logger("Persisted trie from memory database", "nodes", nodes-len(db.dirties)+int(db.flushnodes), "size", storage-db.dirtiesSize+db.flushsize, "time", time.Since(start)+db.flushtime,
   720  		"gcnodes", db.gcnodes, "gcsize", db.gcsize, "gctime", db.gctime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
   721  
   722  	// Reset the garbage collection statistics
   723  	db.gcnodes, db.gcsize, db.gctime = 0, 0, 0
   724  	db.flushnodes, db.flushsize, db.flushtime = 0, 0, 0
   725  
   726  	return nil
   727  }
   728  
   729  // commit is the private locked version of Commit.
   730  func (db *Database) commit(hash common.Hash, batch ethdb.Batch) error {
   731  	// If the node does not exist, it's a previously committed node
   732  	node, ok := db.dirties[hash]
   733  	if !ok {
   734  		return nil
   735  	}
   736  	for _, child := range node.childs() {
   737  		if err := db.commit(child, batch); err != nil {
   738  			return err
   739  		}
   740  	}
   741  	if err := batch.Put(hash[:], node.rlp()); err != nil {
   742  		return err
   743  	}
   744  	// If we've reached an optimal batch size, commit and start over
   745  	if batch.ValueSize() >= ethdb.IdealBatchSize {
   746  		if err := batch.Write(); err != nil {
   747  			return err
   748  		}
   749  		batch.Reset()
   750  	}
   751  	return nil
   752  }
   753  
   754  // uncache is the post-processing step of a commit operation where the already
   755  // persisted trie is removed from the cache. The reason behind the two-phase
   756  // commit is to ensure consistent data availability while moving from memory
   757  // to disk.
   758  func (db *Database) uncache(hash common.Hash) {
   759  	// If the node does not exist, we're done on this path
   760  	node, ok := db.dirties[hash]
   761  	if !ok {
   762  		return
   763  	}
   764  	// Node still exists, remove it from the flush-list
   765  	switch hash {
   766  	case db.oldest:
   767  		db.oldest = node.flushNext
   768  		db.dirties[node.flushNext].flushPrev = common.Hash{}
   769  	case db.newest:
   770  		db.newest = node.flushPrev
   771  		db.dirties[node.flushPrev].flushNext = common.Hash{}
   772  	default:
   773  		db.dirties[node.flushPrev].flushNext = node.flushNext
   774  		db.dirties[node.flushNext].flushPrev = node.flushPrev
   775  	}
   776  	// Uncache the node's subtries and remove the node itself too
   777  	for _, child := range node.childs() {
   778  		db.uncache(child)
   779  	}
   780  	delete(db.dirties, hash)
   781  	db.dirtiesSize -= common.StorageSize(common.HashLength + int(node.size))
   782  }
   783  
   784  // Size returns the current storage size of the memory cache in front of the
   785  // persistent database layer.
   786  func (db *Database) Size() (common.StorageSize, common.StorageSize) {
   787  	db.lock.RLock()
   788  	defer db.lock.RUnlock()
   789  
   790  	// db.dirtiesSize only contains the useful data in the cache, but when reporting
   791  	// the total memory consumption, the maintenance metadata is also needed to be
   792  	// counted. For every useful node, we track 2 extra hashes as the flushlist.
   793  	var flushlistSize = common.StorageSize((len(db.dirties) - 1) * 2 * common.HashLength)
   794  	return db.dirtiesSize + flushlistSize, db.preimagesSize
   795  }
   796  
   797  // verifyIntegrity is a debug method to iterate over the entire trie stored in
   798  // memory and check whether every node is reachable from the meta root. The goal
   799  // is to find any errors that might cause memory leaks and or trie nodes to go
   800  // missing.
   801  //
   802  // This method is extremely CPU and memory intensive, only use when must.
   803  func (db *Database) verifyIntegrity() {
   804  	// Iterate over all the cached nodes and accumulate them into a set
   805  	reachable := map[common.Hash]struct{}{{}: {}}
   806  
   807  	for child := range db.dirties[common.Hash{}].children {
   808  		db.accumulate(child, reachable)
   809  	}
   810  	// Find any unreachable but cached nodes
   811  	unreachable := []string{}
   812  	for hash, node := range db.dirties {
   813  		if _, ok := reachable[hash]; !ok {
   814  			unreachable = append(unreachable, fmt.Sprintf("%x: {Node: %v, Parents: %d, Prev: %x, Next: %x}",
   815  				hash, node.node, node.parents, node.flushPrev, node.flushNext))
   816  		}
   817  	}
   818  	if len(unreachable) != 0 {
   819  		panic(fmt.Sprintf("trie cache memory leak: %v", unreachable))
   820  	}
   821  }
   822  
   823  // accumulate iterates over the trie defined by hash and accumulates all the
   824  // cached children found in memory.
   825  func (db *Database) accumulate(hash common.Hash, reachable map[common.Hash]struct{}) {
   826  	// Mark the node reachable if present in the memory cache
   827  	node, ok := db.dirties[hash]
   828  	if !ok {
   829  		return
   830  	}
   831  	reachable[hash] = struct{}{}
   832  
   833  	// Iterate over all the children and accumulate them too
   834  	for _, child := range node.childs() {
   835  		db.accumulate(child, reachable)
   836  	}
   837  }