aboutsummaryrefslogtreecommitdiffstats
path: root/swarm/storage/pyramid.go
blob: 631ab52b8663c53438e7a569569e6dcd04410627 (plain) (blame)
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
// Copyright 2016 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 storage

import (
    "encoding/binary"
    "errors"
    "io"
    "sync"
    "time"
)

/*
   The main idea of a pyramid chunker is to process the input data without knowing the entire size apriori.
   For this to be achieved, the chunker tree is built from the ground up until the data is exhausted.
   This opens up new aveneus such as easy append and other sort of modifications to the tree thereby avoiding
   duplication of data chunks.


   Below is an example of a two level chunks tree. The leaf chunks are called data chunks and all the above
   chunks are called tree chunks. The tree chunk above data chunks is level 0 and so on until it reaches
   the root tree chunk.



                                            T10                                        <- Tree chunk lvl1
                                            |
                  __________________________|_____________________________
                 /                  |                   |                \
                /                   |                   \                 \
            __T00__             ___T01__           ___T02__           ___T03__         <- Tree chunks lvl 0
           / /     \           / /      \         / /      \         / /      \
          / /       \         / /        \       / /       \        / /        \
         D1 D2 ... D128      D1 D2 ... D128     D1 D2 ... D128     D1 D2 ... D128      <-  Data Chunks


    The split function continuously read the data and creates data chunks and send them to storage.
    When certain no of data chunks are created (defaultBranches), a signal is sent to create a tree
    entry. When the level 0 tree entries reaches certain threshold (defaultBranches), another signal
    is sent to a tree entry one level up.. and so on... until only the data is exhausted AND only one
    tree entry is present in certain level. The key of tree entry is given out as the rootKey of the file.

*/

var (
    errLoadingTreeRootChunk = errors.New("LoadTree Error: Could not load root chunk")
    errLoadingTreeChunk     = errors.New("LoadTree Error: Could not load chunk")
)

const (
    ChunkProcessors       = 8
    DefaultBranches int64 = 128
    splitTimeout          = time.Minute * 5
)

const (
    DataChunk = 0
    TreeChunk = 1
)

type ChunkerParams struct {
    Branches int64
    Hash     string
}

func NewChunkerParams() *ChunkerParams {
    return &ChunkerParams{
        Branches: DefaultBranches,
        Hash:     SHA3Hash,
    }
}

// Entry to create a tree node
type TreeEntry struct {
    level         int
    branchCount   int64
    subtreeSize   uint64
    chunk         []byte
    key           []byte
    index         int  // used in append to indicate the index of existing tree entry
    updatePending bool // indicates if the entry is loaded from existing tree
}

func NewTreeEntry(pyramid *PyramidChunker) *TreeEntry {
    return &TreeEntry{
        level:         0,
        branchCount:   0,
        subtreeSize:   0,
        chunk:         make([]byte, pyramid.chunkSize+8),
        key:           make([]byte, pyramid.hashSize),
        index:         0,
        updatePending: false,
    }
}

// Used by the hash processor to create a data/tree chunk and send to storage
type chunkJob struct {
    key       Key
    chunk     []byte
    size      int64
    parentWg  *sync.WaitGroup
    chunkType int // used to identify the tree related chunks for debugging
    chunkLvl  int // leaf-1 is level 0 and goes upwards until it reaches root
}

type PyramidChunker struct {
    hashFunc    SwarmHasher
    chunkSize   int64
    hashSize    int64
    branches    int64
    workerCount int64
    workerLock  sync.RWMutex
}

func NewPyramidChunker(params *ChunkerParams) (self *PyramidChunker) {
    self = &PyramidChunker{}
    self.hashFunc = MakeHashFunc(params.Hash)
    self.branches = params.Branches
    self.hashSize = int64(self.hashFunc().Size())
    self.chunkSize = self.hashSize * self.branches
    self.workerCount = 0
    return
}

func (self *PyramidChunker) Join(key Key, chunkC chan *Chunk) LazySectionReader {
    return &LazyChunkReader{
        key:       key,
        chunkC:    chunkC,
        chunkSize: self.chunkSize,
        branches:  self.branches,
        hashSize:  self.hashSize,
    }
}

func (self *PyramidChunker) incrementWorkerCount() {
    self.workerLock.Lock()
    defer self.workerLock.Unlock()
    self.workerCount += 1
}

func (self *PyramidChunker) getWorkerCount() int64 {
    self.workerLock.Lock()
    defer self.workerLock.Unlock()
    return self.workerCount
}

func (self *PyramidChunker) decrementWorkerCount() {
    self.workerLock.Lock()
    defer self.workerLock.Unlock()
    self.workerCount -= 1
}

func (self *PyramidChunker) Split(data io.Reader, size int64, chunkC chan *Chunk, storageWG, processorWG *sync.WaitGroup) (Key, error) {
    jobC := make(chan *chunkJob, 2*ChunkProcessors)
    wg := &sync.WaitGroup{}
    errC := make(chan error)
    quitC := make(chan bool)
    rootKey := make([]byte, self.hashSize)
    chunkLevel := make([][]*TreeEntry, self.branches)

    wg.Add(1)
    go self.prepareChunks(false, chunkLevel, data, rootKey, quitC, wg, jobC, processorWG, chunkC, errC, storageWG)

    // closes internal error channel if all subprocesses in the workgroup finished
    go func() {

        // waiting for all chunks to finish
        wg.Wait()

        // if storage waitgroup is non-nil, we wait for storage to finish too
        if storageWG != nil {
            storageWG.Wait()
        }
        //We close errC here because this is passed down to 8 parallel routines underneath.
        // if a error happens in one of them.. that particular routine raises error...
        // once they all complete successfully, the control comes back and we can safely close this here.
        close(errC)
    }()

    defer close(quitC)

    select {
    case err := <-errC:
        if err != nil {
            return nil, err
        }
    case <-time.NewTimer(splitTimeout).C:
    }
    return rootKey, nil

}

func (self *PyramidChunker) Append(key Key, data io.Reader, chunkC chan *Chunk, storageWG, processorWG *sync.WaitGroup) (Key, error) {
    quitC := make(chan bool)
    rootKey := make([]byte, self.hashSize)
    chunkLevel := make([][]*TreeEntry, self.branches)

    // Load the right most unfinished tree chunks in every level
    self.loadTree(chunkLevel, key, chunkC, quitC)

    jobC := make(chan *chunkJob, 2*ChunkProcessors)
    wg := &sync.WaitGroup{}
    errC := make(chan error)

    wg.Add(1)
    go self.prepareChunks(true, chunkLevel, data, rootKey, quitC, wg, jobC, processorWG, chunkC, errC, storageWG)

    // closes internal error channel if all subprocesses in the workgroup finished
    go func() {

        // waiting for all chunks to finish
        wg.Wait()

        // if storage waitgroup is non-nil, we wait for storage to finish too
        if storageWG != nil {
            storageWG.Wait()
        }
        close(errC)
    }()

    defer close(quitC)

    select {
    case err := <-errC:
        if err != nil {
            return nil, err
        }
    case <-time.NewTimer(splitTimeout).C:
    }
    return rootKey, nil

}

func (self *PyramidChunker) processor(id int64, jobC chan *chunkJob, chunkC chan *Chunk, errC chan error, quitC chan bool, swg, wwg *sync.WaitGroup) {
    defer self.decrementWorkerCount()

    hasher := self.hashFunc()
    if wwg != nil {
        defer wwg.Done()
    }
    for {
        select {

        case job, ok := <-jobC:
            if !ok {
                return
            }
            self.processChunk(id, hasher, job, chunkC, swg)
        case <-quitC:
            return
        }
    }
}

func (self *PyramidChunker) processChunk(id int64, hasher SwarmHash, job *chunkJob, chunkC chan *Chunk, swg *sync.WaitGroup) {
    hasher.ResetWithLength(job.chunk[:8]) // 8 bytes of length
    hasher.Write(job.chunk[8:])           // minus 8 []byte length
    h := hasher.Sum(nil)

    newChunk := &Chunk{
        Key:   h,
        SData: job.chunk,
        Size:  job.size,
        wg:    swg,
    }

    // report hash of this chunk one level up (keys corresponds to the proper subslice of the parent chunk)
    copy(job.key, h)

    // send off new chunk to storage
    if chunkC != nil {
        if swg != nil {
            swg.Add(1)
        }
    }
    job.parentWg.Done()

    if chunkC != nil {
        chunkC <- newChunk
    }
}

func (self *PyramidChunker) loadTree(chunkLevel [][]*TreeEntry, key Key, chunkC chan *Chunk, quitC chan bool) error {
    // Get the root chunk to get the total size
    chunk := retrieve(key, chunkC, quitC)
    if chunk == nil {
        return errLoadingTreeRootChunk
    }

    //if data size is less than a chunk... add a parent with update as pending
    if chunk.Size <= self.chunkSize {
        newEntry := &TreeEntry{
            level:         0,
            branchCount:   1,
            subtreeSize:   uint64(chunk.Size),
            chunk:         make([]byte, self.chunkSize+8),
            key:           make([]byte, self.hashSize),
            index:         0,
            updatePending: true,
        }
        copy(newEntry.chunk[8:], chunk.Key)
        chunkLevel[0] = append(chunkLevel[0], newEntry)
        return nil
    }

    var treeSize int64
    var depth int
    treeSize = self.chunkSize
    for ; treeSize < chunk.Size; treeSize *= self.branches {
        depth++
    }

    // Add the root chunk entry
    branchCount := int64(len(chunk.SData)-8) / self.hashSize
    newEntry := &TreeEntry{
        level:         int(depth - 1),
        branchCount:   branchCount,
        subtreeSize:   uint64(chunk.Size),
        chunk:         chunk.SData,
        key:           key,
        index:         0,
        updatePending: true,
    }
    chunkLevel[depth-1] = append(chunkLevel[depth-1], newEntry)

    // Add the rest of the tree
    for lvl := (depth - 1); lvl >= 1; lvl-- {

        //TODO(jmozah): instead of loading finished branches and then trim in the end,
        //avoid loading them in the first place
        for _, ent := range chunkLevel[lvl] {
            branchCount = int64(len(ent.chunk)-8) / self.hashSize
            for i := int64(0); i < branchCount; i++ {
                key := ent.chunk[8+(i*self.hashSize) : 8+((i+1)*self.hashSize)]
                newChunk := retrieve(key, chunkC, quitC)
                if newChunk == nil {
                    return errLoadingTreeChunk
                }
                bewBranchCount := int64(len(newChunk.SData)-8) / self.hashSize
                newEntry := &TreeEntry{
                    level:         int(lvl - 1),
                    branchCount:   bewBranchCount,
                    subtreeSize:   uint64(newChunk.Size),
                    chunk:         newChunk.SData,
                    key:           key,
                    index:         0,
                    updatePending: true,
                }
                chunkLevel[lvl-1] = append(chunkLevel[lvl-1], newEntry)

            }

            // We need to get only the right most unfinished branch.. so trim all finished branches
            if int64(len(chunkLevel[lvl-1])) >= self.branches {
                chunkLevel[lvl-1] = nil
            }
        }
    }

    return nil
}

func (self *PyramidChunker) prepareChunks(isAppend bool, chunkLevel [][]*TreeEntry, data io.Reader, rootKey []byte, quitC chan bool, wg *sync.WaitGroup, jobC chan *chunkJob, processorWG *sync.WaitGroup, chunkC chan *Chunk, errC chan error, storageWG *sync.WaitGroup) {
    defer wg.Done()

    chunkWG := &sync.WaitGroup{}
    totalDataSize := 0

    // processorWG keeps track of workers spawned for hashing chunks
    if processorWG != nil {
        processorWG.Add(1)
    }

    self.incrementWorkerCount()
    go self.processor(self.workerCount, jobC, chunkC, errC, quitC, storageWG, processorWG)

    parent := NewTreeEntry(self)
    var unFinishedChunk *Chunk

    if isAppend == true && len(chunkLevel[0]) != 0 {

        lastIndex := len(chunkLevel[0]) - 1
        ent := chunkLevel[0][lastIndex]

        if ent.branchCount < self.branches {
            parent = &TreeEntry{
                level:         0,
                branchCount:   ent.branchCount,
                subtreeSize:   ent.subtreeSize,
                chunk:         ent.chunk,
                key:           ent.key,
                index:         lastIndex,
                updatePending: true,
            }

            lastBranch := parent.branchCount - 1
            lastKey := parent.chunk[8+lastBranch*self.hashSize : 8+(lastBranch+1)*self.hashSize]

            unFinishedChunk = retrieve(lastKey, chunkC, quitC)
            if unFinishedChunk.Size < self.chunkSize {

                parent.subtreeSize = parent.subtreeSize - uint64(unFinishedChunk.Size)
                parent.branchCount = parent.branchCount - 1
            } else {
                unFinishedChunk = nil
            }
        }
    }

    for index := 0; ; index++ {

        var n int
        var err error
        chunkData := make([]byte, self.chunkSize+8)
        if unFinishedChunk != nil {
            copy(chunkData, unFinishedChunk.SData)
            n, err = data.Read(chunkData[8+unFinishedChunk.Size:])
            n += int(unFinishedChunk.Size)
            unFinishedChunk = nil
        } else {
            n, err = data.Read(chunkData[8:])
        }

        totalDataSize += n
        if err != nil {
            if err == io.EOF || err == io.ErrUnexpectedEOF {
                if parent.branchCount == 1 {
                    // Data is exactly one chunk.. pick the last chunk key as root
                    chunkWG.Wait()
                    lastChunksKey := parent.chunk[8 : 8+self.hashSize]
                    copy(rootKey, lastChunksKey)
                    break
                }
            } else {
                close(quitC)
                break
            }
        }

        // Data ended in chunk boundary.. just signal to start bulding tree
        if n == 0 {
            self.buildTree(isAppend, chunkLevel, parent, chunkWG, jobC, quitC, true, rootKey)
            break
        } else {

            pkey := self.enqueueDataChunk(chunkData, uint64(n), parent, chunkWG, jobC, quitC)

            // update tree related parent data structures
            parent.subtreeSize += uint64(n)
            parent.branchCount++

            // Data got exhausted... signal to send any parent tree related chunks
            if int64(n) < self.chunkSize {

                // only one data chunk .. so dont add any parent chunk
                if parent.branchCount <= 1 {
                    chunkWG.Wait()
                    copy(rootKey, pkey)
                    break
                }

                self.buildTree(isAppend, chunkLevel, parent, chunkWG, jobC, quitC, true, rootKey)
                break
            }

            if parent.branchCount == self.branches {
                self.buildTree(isAppend, chunkLevel, parent, chunkWG, jobC, quitC, false, rootKey)
                parent = NewTreeEntry(self)
            }

        }

        workers := self.getWorkerCount()
        if int64(len(jobC)) > workers && workers < ChunkProcessors {
            if processorWG != nil {
                processorWG.Add(1)
            }
            self.incrementWorkerCount()
            go self.processor(self.workerCount, jobC, chunkC, errC, quitC, storageWG, processorWG)
        }

    }

}

func (self *PyramidChunker) buildTree(isAppend bool, chunkLevel [][]*TreeEntry, ent *TreeEntry, chunkWG *sync.WaitGroup, jobC chan *chunkJob, quitC chan bool, last bool, rootKey []byte) {
    chunkWG.Wait()
    self.enqueueTreeChunk(chunkLevel, ent, chunkWG, jobC, quitC, last)

    compress := false
    endLvl := self.branches
    for lvl := int64(0); lvl < self.branches; lvl++ {
        lvlCount := int64(len(chunkLevel[lvl]))
        if lvlCount >= self.branches {
            endLvl = lvl + 1
            compress = true
            break
        }
    }

    if compress == false && last == false {
        return
    }

    // Wait for all the keys to be processed before compressing the tree
    chunkWG.Wait()

    for lvl := int64(ent.level); lvl < endLvl; lvl++ {

        lvlCount := int64(len(chunkLevel[lvl]))
        if lvlCount == 1 && last == true {
            copy(rootKey, chunkLevel[lvl][0].key)
            return
        }

        for startCount := int64(0); startCount < lvlCount; startCount += self.branches {

            endCount := startCount + self.branches
            if endCount > lvlCount {
                endCount = lvlCount
            }

            var nextLvlCount int64
            var tempEntry *TreeEntry
            if len(chunkLevel[lvl+1]) > 0 {
                nextLvlCount = int64(len(chunkLevel[lvl+1]) - 1)
                tempEntry = chunkLevel[lvl+1][nextLvlCount]
            }
            if isAppend == true && tempEntry != nil && tempEntry.updatePending == true {
                updateEntry := &TreeEntry{
                    level:         int(lvl + 1),
                    branchCount:   0,
                    subtreeSize:   0,
                    chunk:         make([]byte, self.chunkSize+8),
                    key:           make([]byte, self.hashSize),
                    index:         int(nextLvlCount),
                    updatePending: true,
                }
                for index := int64(0); index < lvlCount; index++ {
                    updateEntry.branchCount++
                    updateEntry.subtreeSize += chunkLevel[lvl][index].subtreeSize
                    copy(updateEntry.chunk[8+(index*self.hashSize):8+((index+1)*self.hashSize)], chunkLevel[lvl][index].key[:self.hashSize])
                }

                self.enqueueTreeChunk(chunkLevel, updateEntry, chunkWG, jobC, quitC, last)

            } else {

                noOfBranches := endCount - startCount
                newEntry := &TreeEntry{
                    level:         int(lvl + 1),
                    branchCount:   noOfBranches,
                    subtreeSize:   0,
                    chunk:         make([]byte, (noOfBranches*self.hashSize)+8),
                    key:           make([]byte, self.hashSize),
                    index:         int(nextLvlCount),
                    updatePending: false,
                }

                index := int64(0)
                for i := startCount; i < endCount; i++ {
                    entry := chunkLevel[lvl][i]
                    newEntry.subtreeSize += entry.subtreeSize
                    copy(newEntry.chunk[8+(index*self.hashSize):8+((index+1)*self.hashSize)], entry.key[:self.hashSize])
                    index++
                }

                self.enqueueTreeChunk(chunkLevel, newEntry, chunkWG, jobC, quitC, last)

            }

        }

        if isAppend == false {
            chunkWG.Wait()
            if compress == true {
                chunkLevel[lvl] = nil
            }
        }
    }

}

func (self *PyramidChunker) enqueueTreeChunk(chunkLevel [][]*TreeEntry, ent *TreeEntry, chunkWG *sync.WaitGroup, jobC chan *chunkJob, quitC chan bool, last bool) {
    if ent != nil {

        // wait for data chunks to get over before processing the tree chunk
        if last == true {
            chunkWG.Wait()
        }

        binary.LittleEndian.PutUint64(ent.chunk[:8], ent.subtreeSize)
        ent.key = make([]byte, self.hashSize)
        chunkWG.Add(1)
        select {
        case jobC <- &chunkJob{ent.key, ent.chunk[:ent.branchCount*self.hashSize+8], int64(ent.subtreeSize), chunkWG, TreeChunk, 0}:
        case <-quitC:
        }

        // Update or append based on weather it is a new entry or being reused
        if ent.updatePending == true {
            chunkWG.Wait()
            chunkLevel[ent.level][ent.index] = ent
        } else {
            chunkLevel[ent.level] = append(chunkLevel[ent.level], ent)
        }

    }
}

func (self *PyramidChunker) enqueueDataChunk(chunkData []byte, size uint64, parent *TreeEntry, chunkWG *sync.WaitGroup, jobC chan *chunkJob, quitC chan bool) Key {
    binary.LittleEndian.PutUint64(chunkData[:8], size)
    pkey := parent.chunk[8+parent.branchCount*self.hashSize : 8+(parent.branchCount+1)*self.hashSize]

    chunkWG.Add(1)
    select {
    case jobC <- &chunkJob{pkey, chunkData[:size+8], int64(size), chunkWG, DataChunk, -1}:
    case <-quitC:
    }

    return pkey

}