aboutsummaryrefslogtreecommitdiffstats
path: root/swarm/storage/chunker.go
blob: b9b5022734c023c7e3ec1e7b2b80c47b5f3632d9 (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
// 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 (
    "context"
    "encoding/binary"
    "errors"
    "fmt"
    "io"
    "sync"
    "time"

    "github.com/ethereum/go-ethereum/metrics"
    "github.com/ethereum/go-ethereum/swarm/log"
    "github.com/ethereum/go-ethereum/swarm/spancontext"
    opentracing "github.com/opentracing/opentracing-go"
    olog "github.com/opentracing/opentracing-go/log"
)

/*
The distributed storage implemented in this package requires fix sized chunks of content.

Chunker is the interface to a component that is responsible for disassembling and assembling larger data.

TreeChunker implements a Chunker based on a tree structure defined as follows:

1 each node in the tree including the root and other branching nodes are stored as a chunk.

2 branching nodes encode data contents that includes the size of the dataslice covered by its entire subtree under the node as well as the hash keys of all its children :
data_{i} := size(subtree_{i}) || key_{j} || key_{j+1} .... || key_{j+n-1}

3 Leaf nodes encode an actual subslice of the input data.

4 if data size is not more than maximum chunksize, the data is stored in a single chunk
  key = hash(int64(size) + data)

5 if data size is more than chunksize*branches^l, but no more than chunksize*
  branches^(l+1), the data vector is split into slices of chunksize*
  branches^l length (except the last one).
  key = hash(int64(size) + key(slice0) + key(slice1) + ...)

 The underlying hash function is configurable
*/

/*
Tree chunker is a concrete implementation of data chunking.
This chunker works in a simple way, it builds a tree out of the document so that each node either represents a chunk of real data or a chunk of data representing an branching non-leaf node of the tree. In particular each such non-leaf chunk will represent is a concatenation of the hash of its respective children. This scheme simultaneously guarantees data integrity as well as self addressing. Abstract nodes are transparent since their represented size component is strictly greater than their maximum data size, since they encode a subtree.

If all is well it is possible to implement this by simply composing readers so that no extra allocation or buffering is necessary for the data splitting and joining. This means that in principle there can be direct IO between : memory, file system, network socket (bzz peers storage request is read from the socket). In practice there may be need for several stages of internal buffering.
The hashing itself does use extra copies and allocation though, since it does need it.
*/

var (
    errAppendOppNotSuported = errors.New("Append operation not supported")
    errOperationTimedOut    = errors.New("operation timed out")
)

const (
    DefaultChunkSize int64 = 4096
)

type ChunkerParams struct {
    chunkSize int64
    hashSize  int64
}

type SplitterParams struct {
    ChunkerParams
    reader io.Reader
    putter Putter
    addr   Address
}

type TreeSplitterParams struct {
    SplitterParams
    size int64
}

type JoinerParams struct {
    ChunkerParams
    addr   Address
    getter Getter
    // TODO: there is a bug, so depth can only be 0 today, see: https://github.com/ethersphere/go-ethereum/issues/344
    depth int
    ctx   context.Context
}

type TreeChunker struct {
    ctx context.Context

    branches int64
    hashFunc SwarmHasher
    dataSize int64
    data     io.Reader
    // calculated
    addr        Address
    depth       int
    hashSize    int64        // self.hashFunc.New().Size()
    chunkSize   int64        // hashSize* branches
    workerCount int64        // the number of worker routines used
    workerLock  sync.RWMutex // lock for the worker count
    jobC        chan *hashJob
    wg          *sync.WaitGroup
    putter      Putter
    getter      Getter
    errC        chan error
    quitC       chan bool
}

/*
    Join reconstructs original content based on a root key.
    When joining, the caller gets returned a Lazy SectionReader, which is
    seekable and implements on-demand fetching of chunks as and where it is read.
    New chunks to retrieve are coming from the getter, which the caller provides.
    If an error is encountered during joining, it appears as a reader error.
    The SectionReader.
    As a result, partial reads from a document are possible even if other parts
    are corrupt or lost.
    The chunks are not meant to be validated by the chunker when joining. This
    is because it is left to the DPA to decide which sources are trusted.
*/
func TreeJoin(ctx context.Context, addr Address, getter Getter, depth int) *LazyChunkReader {
    jp := &JoinerParams{
        ChunkerParams: ChunkerParams{
            chunkSize: DefaultChunkSize,
            hashSize:  int64(len(addr)),
        },
        addr:   addr,
        getter: getter,
        depth:  depth,
        ctx:    ctx,
    }

    return NewTreeJoiner(jp).Join(ctx)
}

/*
    When splitting, data is given as a SectionReader, and the key is a hashSize long byte slice (Key), the root hash of the entire content will fill this once processing finishes.
    New chunks to store are store using the putter which the caller provides.
*/
func TreeSplit(ctx context.Context, data io.Reader, size int64, putter Putter) (k Address, wait func(context.Context) error, err error) {
    tsp := &TreeSplitterParams{
        SplitterParams: SplitterParams{
            ChunkerParams: ChunkerParams{
                chunkSize: DefaultChunkSize,
                hashSize:  putter.RefSize(),
            },
            reader: data,
            putter: putter,
        },
        size: size,
    }
    return NewTreeSplitter(tsp).Split(ctx)
}

func NewTreeJoiner(params *JoinerParams) *TreeChunker {
    tc := &TreeChunker{}
    tc.hashSize = params.hashSize
    tc.branches = params.chunkSize / params.hashSize
    tc.addr = params.addr
    tc.getter = params.getter
    tc.depth = params.depth
    tc.chunkSize = params.chunkSize
    tc.workerCount = 0
    tc.jobC = make(chan *hashJob, 2*ChunkProcessors)
    tc.wg = &sync.WaitGroup{}
    tc.errC = make(chan error)
    tc.quitC = make(chan bool)

    tc.ctx = params.ctx

    return tc
}

func NewTreeSplitter(params *TreeSplitterParams) *TreeChunker {
    tc := &TreeChunker{}
    tc.data = params.reader
    tc.dataSize = params.size
    tc.hashSize = params.hashSize
    tc.branches = params.chunkSize / params.hashSize
    tc.addr = params.addr
    tc.chunkSize = params.chunkSize
    tc.putter = params.putter
    tc.workerCount = 0
    tc.jobC = make(chan *hashJob, 2*ChunkProcessors)
    tc.wg = &sync.WaitGroup{}
    tc.errC = make(chan error)
    tc.quitC = make(chan bool)

    return tc
}

// String() for pretty printing
func (c *Chunk) String() string {
    return fmt.Sprintf("Key: %v TreeSize: %v Chunksize: %v", c.Addr.Log(), c.Size, len(c.SData))
}

type hashJob struct {
    key      Address
    chunk    []byte
    size     int64
    parentWg *sync.WaitGroup
}

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

func (tc *TreeChunker) getWorkerCount() int64 {
    tc.workerLock.RLock()
    defer tc.workerLock.RUnlock()
    return tc.workerCount
}

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

func (tc *TreeChunker) Split(ctx context.Context) (k Address, wait func(context.Context) error, err error) {
    if tc.chunkSize <= 0 {
        panic("chunker must be initialised")
    }

    tc.runWorker()

    depth := 0
    treeSize := tc.chunkSize

    // takes lowest depth such that chunksize*HashCount^(depth+1) > size
    // power series, will find the order of magnitude of the data size in base hashCount or numbers of levels of branching in the resulting tree.
    for ; treeSize < tc.dataSize; treeSize *= tc.branches {
        depth++
    }

    key := make([]byte, tc.hashSize)
    // this waitgroup member is released after the root hash is calculated
    tc.wg.Add(1)
    //launch actual recursive function passing the waitgroups
    go tc.split(depth, treeSize/tc.branches, key, tc.dataSize, tc.wg)

    // closes internal error channel if all subprocesses in the workgroup finished
    go func() {
        // waiting for all threads to finish
        tc.wg.Wait()
        close(tc.errC)
    }()

    defer close(tc.quitC)
    defer tc.putter.Close()
    select {
    case err := <-tc.errC:
        if err != nil {
            return nil, nil, err
        }
    case <-time.NewTimer(splitTimeout).C:
        return nil, nil, errOperationTimedOut
    }

    return key, tc.putter.Wait, nil
}

func (tc *TreeChunker) split(depth int, treeSize int64, addr Address, size int64, parentWg *sync.WaitGroup) {

    //

    for depth > 0 && size < treeSize {
        treeSize /= tc.branches
        depth--
    }

    if depth == 0 {
        // leaf nodes -> content chunks
        chunkData := make([]byte, size+8)
        binary.LittleEndian.PutUint64(chunkData[0:8], uint64(size))
        var readBytes int64
        for readBytes < size {
            n, err := tc.data.Read(chunkData[8+readBytes:])
            readBytes += int64(n)
            if err != nil && !(err == io.EOF && readBytes == size) {
                tc.errC <- err
                return
            }
        }
        select {
        case tc.jobC <- &hashJob{addr, chunkData, size, parentWg}:
        case <-tc.quitC:
        }
        return
    }
    // dept > 0
    // intermediate chunk containing child nodes hashes
    branchCnt := (size + treeSize - 1) / treeSize

    var chunk = make([]byte, branchCnt*tc.hashSize+8)
    var pos, i int64

    binary.LittleEndian.PutUint64(chunk[0:8], uint64(size))

    childrenWg := &sync.WaitGroup{}
    var secSize int64
    for i < branchCnt {
        // the last item can have shorter data
        if size-pos < treeSize {
            secSize = size - pos
        } else {
            secSize = treeSize
        }
        // the hash of that data
        subTreeKey := chunk[8+i*tc.hashSize : 8+(i+1)*tc.hashSize]

        childrenWg.Add(1)
        tc.split(depth-1, treeSize/tc.branches, subTreeKey, secSize, childrenWg)

        i++
        pos += treeSize
    }
    // wait for all the children to complete calculating their hashes and copying them onto sections of the chunk
    // parentWg.Add(1)
    // go func() {
    childrenWg.Wait()

    worker := tc.getWorkerCount()
    if int64(len(tc.jobC)) > worker && worker < ChunkProcessors {
        tc.runWorker()

    }
    select {
    case tc.jobC <- &hashJob{addr, chunk, size, parentWg}:
    case <-tc.quitC:
    }
}

func (tc *TreeChunker) runWorker() {
    tc.incrementWorkerCount()
    go func() {
        defer tc.decrementWorkerCount()
        for {
            select {

            case job, ok := <-tc.jobC:
                if !ok {
                    return
                }

                h, err := tc.putter.Put(tc.ctx, job.chunk)
                if err != nil {
                    tc.errC <- err
                    return
                }
                copy(job.key, h)
                job.parentWg.Done()
            case <-tc.quitC:
                return
            }
        }
    }()
}

func (tc *TreeChunker) Append() (Address, func(), error) {
    return nil, nil, errAppendOppNotSuported
}

// LazyChunkReader implements LazySectionReader
type LazyChunkReader struct {
    Ctx       context.Context
    key       Address // root key
    chunkData ChunkData
    off       int64 // offset
    chunkSize int64 // inherit from chunker
    branches  int64 // inherit from chunker
    hashSize  int64 // inherit from chunker
    depth     int
    getter    Getter
}

func (tc *TreeChunker) Join(ctx context.Context) *LazyChunkReader {
    return &LazyChunkReader{
        key:       tc.addr,
        chunkSize: tc.chunkSize,
        branches:  tc.branches,
        hashSize:  tc.hashSize,
        depth:     tc.depth,
        getter:    tc.getter,
        Ctx:       tc.ctx,
    }
}

func (r *LazyChunkReader) Context() context.Context {
    return r.Ctx
}

// Size is meant to be called on the LazySectionReader
func (r *LazyChunkReader) Size(ctx context.Context, quitC chan bool) (n int64, err error) {
    metrics.GetOrRegisterCounter("lazychunkreader.size", nil).Inc(1)

    var sp opentracing.Span
    var cctx context.Context
    cctx, sp = spancontext.StartSpan(
        ctx,
        "lcr.size")
    defer sp.Finish()

    log.Debug("lazychunkreader.size", "key", r.key)
    if r.chunkData == nil {
        chunkData, err := r.getter.Get(cctx, Reference(r.key))
        if err != nil {
            return 0, err
        }
        if chunkData == nil {
            select {
            case <-quitC:
                return 0, errors.New("aborted")
            default:
                return 0, fmt.Errorf("root chunk not found for %v", r.key.Hex())
            }
        }
        r.chunkData = chunkData
    }
    return r.chunkData.Size(), nil
}

// read at can be called numerous times
// concurrent reads are allowed
// Size() needs to be called synchronously on the LazyChunkReader first
func (r *LazyChunkReader) ReadAt(b []byte, off int64) (read int, err error) {
    metrics.GetOrRegisterCounter("lazychunkreader.readat", nil).Inc(1)

    var sp opentracing.Span
    var cctx context.Context
    cctx, sp = spancontext.StartSpan(
        r.Ctx,
        "lcr.read")
    defer sp.Finish()

    defer func() {
        sp.LogFields(
            olog.Int("off", int(off)),
            olog.Int("read", read))
    }()

    // this is correct, a swarm doc cannot be zero length, so no EOF is expected
    if len(b) == 0 {
        return 0, nil
    }
    quitC := make(chan bool)
    size, err := r.Size(cctx, quitC)
    if err != nil {
        log.Error("lazychunkreader.readat.size", "size", size, "err", err)
        return 0, err
    }

    errC := make(chan error)

    // }
    var treeSize int64
    var depth int
    // calculate depth and max treeSize
    treeSize = r.chunkSize
    for ; treeSize < size; treeSize *= r.branches {
        depth++
    }
    wg := sync.WaitGroup{}
    length := int64(len(b))
    for d := 0; d < r.depth; d++ {
        off *= r.chunkSize
        length *= r.chunkSize
    }
    wg.Add(1)
    go r.join(cctx, b, off, off+length, depth, treeSize/r.branches, r.chunkData, &wg, errC, quitC)
    go func() {
        wg.Wait()
        close(errC)
    }()

    err = <-errC
    if err != nil {
        log.Error("lazychunkreader.readat.errc", "err", err)
        close(quitC)
        return 0, err
    }
    if off+int64(len(b)) >= size {
        return int(size - off), io.EOF
    }
    return len(b), nil
}

func (r *LazyChunkReader) join(ctx context.Context, b []byte, off int64, eoff int64, depth int, treeSize int64, chunkData ChunkData, parentWg *sync.WaitGroup, errC chan error, quitC chan bool) {
    defer parentWg.Done()
    // find appropriate block level
    for chunkData.Size() < treeSize && depth > r.depth {
        treeSize /= r.branches
        depth--
    }

    // leaf chunk found
    if depth == r.depth {
        extra := 8 + eoff - int64(len(chunkData))
        if extra > 0 {
            eoff -= extra
        }
        copy(b, chunkData[8+off:8+eoff])
        return // simply give back the chunks reader for content chunks
    }

    // subtree
    start := off / treeSize
    end := (eoff + treeSize - 1) / treeSize

    // last non-leaf chunk can be shorter than default chunk size, let's not read it further then its end
    currentBranches := int64(len(chunkData)-8) / r.hashSize
    if end > currentBranches {
        end = currentBranches
    }

    wg := &sync.WaitGroup{}
    defer wg.Wait()
    for i := start; i < end; i++ {
        soff := i * treeSize
        roff := soff
        seoff := soff + treeSize

        if soff < off {
            soff = off
        }
        if seoff > eoff {
            seoff = eoff
        }
        if depth > 1 {
            wg.Wait()
        }
        wg.Add(1)
        go func(j int64) {
            childKey := chunkData[8+j*r.hashSize : 8+(j+1)*r.hashSize]
            chunkData, err := r.getter.Get(ctx, Reference(childKey))
            if err != nil {
                log.Error("lazychunkreader.join", "key", fmt.Sprintf("%x", childKey), "err", err)
                select {
                case errC <- fmt.Errorf("chunk %v-%v not found; key: %s", off, off+treeSize, fmt.Sprintf("%x", childKey)):
                case <-quitC:
                }
                return
            }
            if l := len(chunkData); l < 9 {
                select {
                case errC <- fmt.Errorf("chunk %v-%v incomplete; key: %s, data length %v", off, off+treeSize, fmt.Sprintf("%x", childKey), l):
                case <-quitC:
                }
                return
            }
            if soff < off {
                soff = off
            }
            r.join(ctx, b[soff-off:seoff-off], soff-roff, seoff-roff, depth-1, treeSize/r.branches, chunkData, wg, errC, quitC)
        }(i)
    } //for
}

// Read keeps a cursor so cannot be called simulateously, see ReadAt
func (r *LazyChunkReader) Read(b []byte) (read int, err error) {
    log.Debug("lazychunkreader.read", "key", r.key)
    metrics.GetOrRegisterCounter("lazychunkreader.read", nil).Inc(1)

    read, err = r.ReadAt(b, r.off)
    if err != nil && err != io.EOF {
        log.Error("lazychunkreader.readat", "read", read, "err", err)
        metrics.GetOrRegisterCounter("lazychunkreader.read.err", nil).Inc(1)
    }

    metrics.GetOrRegisterCounter("lazychunkreader.read.bytes", nil).Inc(int64(read))

    r.off += int64(read)
    return
}

// completely analogous to standard SectionReader implementation
var errWhence = errors.New("Seek: invalid whence")
var errOffset = errors.New("Seek: invalid offset")

func (r *LazyChunkReader) Seek(offset int64, whence int) (int64, error) {
    log.Debug("lazychunkreader.seek", "key", r.key, "offset", offset)
    switch whence {
    default:
        return 0, errWhence
    case 0:
        offset += 0
    case 1:
        offset += r.off
    case 2:
        if r.chunkData == nil { //seek from the end requires rootchunk for size. call Size first
            _, err := r.Size(context.TODO(), nil)
            if err != nil {
                return 0, fmt.Errorf("can't get size: %v", err)
            }
        }
        offset += r.chunkData.Size()
    }

    if offset < 0 {
        return 0, errOffset
    }
    r.off = offset
    return offset, nil
}