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