aboutsummaryrefslogtreecommitdiffstats
path: root/eth/downloader/downloader.go
blob: 1707e1395772c0141bc73d6ae1bcd31835eed9a9 (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
package downloader

import (
    "math"
    "math/big"
    "sync"
    "sync/atomic"
    "time"

    "github.com/ethereum/go-ethereum/common"
    "github.com/ethereum/go-ethereum/core"
    "github.com/ethereum/go-ethereum/core/types"
    "github.com/ethereum/go-ethereum/logger"
    "github.com/ethereum/go-ethereum/logger/glog"
    "gopkg.in/fatih/set.v0"
)

const (
    maxBlockFetch       = 256 // Amount of max blocks to be fetched per chunk
    minDesiredPeerCount = 3   // Amount of peers desired to start syncing
)

type hashCheckFn func(common.Hash) bool
type chainInsertFn func(types.Blocks) error
type hashIterFn func() (common.Hash, error)
type currentTdFn func() *big.Int

type Downloader struct {
    mu    sync.RWMutex
    queue *queue
    peers peers

    // Callbacks
    hasBlock    hashCheckFn
    insertChain chainInsertFn
    currentTd   currentTdFn

    // Status
    fetchingHashes    int32
    downloadingBlocks int32
    processingBlocks  int32

    // Channels
    newPeerCh chan *peer
    syncCh    chan syncPack
    HashCh    chan []common.Hash
    blockCh   chan blockPack
    quit      chan struct{}
}

type blockPack struct {
    peerId string
    blocks []*types.Block
}

type syncPack struct {
    peer          *peer
    hash          common.Hash
    ignoreInitial bool
}

func New(hasBlock hashCheckFn, insertChain chainInsertFn, currentTd currentTdFn) *Downloader {
    downloader := &Downloader{
        queue:       newqueue(),
        peers:       make(peers),
        hasBlock:    hasBlock,
        insertChain: insertChain,
        currentTd:   currentTd,
        newPeerCh:   make(chan *peer, 1),
        syncCh:      make(chan syncPack, 1),
        HashCh:      make(chan []common.Hash, 1),
        blockCh:     make(chan blockPack, 1),
        quit:        make(chan struct{}),
    }
    go downloader.peerHandler()
    go downloader.update()

    return downloader
}

func (d *Downloader) RegisterPeer(id string, td *big.Int, hash common.Hash, getHashes hashFetcherFn, getBlocks blockFetcherFn) error {
    d.mu.Lock()
    defer d.mu.Unlock()

    glog.V(logger.Detail).Infoln("Register peer", id)

    // Create a new peer and add it to the list of known peers
    peer := newPeer(id, td, hash, getHashes, getBlocks)
    // add peer to our peer set
    d.peers[id] = peer
    // broadcast new peer
    d.newPeerCh <- peer

    return nil
}

func (d *Downloader) UnregisterPeer(id string) {
    d.mu.Lock()
    defer d.mu.Unlock()

    glog.V(logger.Detail).Infoln("Unregister peer", id)

    delete(d.peers, id)
}

func (d *Downloader) peerHandler() {
    // itimer is used to determine when to start ignoring `minDesiredPeerCount`
    //itimer := time.NewTicker(5 * time.Second)
    itimer := time.NewTimer(5 * time.Second)
out:
    for {
        select {
        case <-d.newPeerCh:
            itimer.Stop()
            // Meet the `minDesiredPeerCount` before we select our best peer
            if len(d.peers) < minDesiredPeerCount {
                break
            }
            d.selectPeer(d.peers.bestPeer())
        case <-itimer.C:
            // The timer will make sure that the downloader keeps an active state
            // in which it attempts to always check the network for highest td peers
            d.selectPeer(d.peers.bestPeer())
        case <-d.quit:
            break out
        }
    }
}

func (d *Downloader) selectPeer(p *peer) {
    // Make sure it's doing neither. Once done we can restart the
    // downloading process if the TD is higher. For now just get on
    // with whatever is going on. This prevents unecessary switching.
    if !(d.isFetchingHashes() || d.isDownloadingBlocks() || d.isProcessing()) {
        // selected peer must be better than our own
        // XXX we also check the peer's recent hash to make sure we
        // don't have it. Some peers report (i think) incorrect TD.
        if p.td.Cmp(d.currentTd()) <= 0 || d.hasBlock(p.recentHash) {
            return
        }

        glog.V(logger.Detail).Infoln("New peer with highest TD =", p.td)
        d.syncCh <- syncPack{p, p.recentHash, false}
    }
}

func (d *Downloader) update() {
out:
    for {
        select {
        case sync := <-d.syncCh:
            selectedPeer := sync.peer
            glog.V(logger.Detail).Infoln("Synchronising with the network using:", selectedPeer.id)
            // Start the fetcher. This will block the update entirely
            // interupts need to be send to the appropriate channels
            // respectively.
            if err := d.startFetchingHashes(selectedPeer, sync.hash, sync.ignoreInitial); err != nil {
                // handle error
                glog.V(logger.Debug).Infoln("Error fetching hashes:", err)
                // XXX Reset
                break
            }

            // Start fetching blocks in paralel. The strategy is simple
            // take any available peers, seserve a chunk for each peer available,
            // let the peer deliver the chunkn and periodically check if a peer
            // has timedout. When done downloading, process blocks.
            if err := d.startFetchingBlocks(selectedPeer); err != nil {
                glog.V(logger.Debug).Infoln("Error downloading blocks:", err)
                // XXX reset
                break
            }

            glog.V(logger.Detail).Infoln("Sync completed")

            d.process()
        case <-d.quit:
            break out
        }
    }
}

// XXX Make synchronous
func (d *Downloader) startFetchingHashes(p *peer, hash common.Hash, ignoreInitial bool) error {
    glog.V(logger.Debug).Infof("Downloading hashes (%x) from %s", hash.Bytes()[:4], p.id)

    start := time.Now()

    // We ignore the initial hash in some cases (e.g. we received a block without it's parent)
    // In such circumstances we don't need to download the block so don't add it to the queue.
    if !ignoreInitial {
        // Add the hash to the queue first
        d.queue.hashPool.Add(hash)
    }

    // Get the first batch of hashes
    p.getHashes(hash)
    atomic.StoreInt32(&d.fetchingHashes, 1)

out:
    for {
        select {
        case hashes := <-d.HashCh:
            var done bool // determines whether we're done fetching hashes (i.e. common hash found)
            hashSet := set.New()
            for _, hash := range hashes {
                if d.hasBlock(hash) {
                    glog.V(logger.Debug).Infof("Found common hash %x\n", hash[:4])

                    done = true
                    break
                }

                hashSet.Add(hash)
            }
            d.queue.put(hashSet)

            // Add hashes to the chunk set
            // Check if we're done fetching
            if !done && len(hashes) > 0 {
                //fmt.Println("re-fetch. current =", d.queue.hashPool.Size())
                // Get the next set of hashes
                p.getHashes(hashes[len(hashes)-1])
                atomic.StoreInt32(&d.fetchingHashes, 1)
            } else {
                atomic.StoreInt32(&d.fetchingHashes, 0)
                break out
            }
        }
    }
    glog.V(logger.Detail).Infof("Downloaded hashes (%d). Took %v\n", d.queue.hashPool.Size(), time.Since(start))

    return nil
}

func (d *Downloader) startFetchingBlocks(p *peer) error {
    glog.V(logger.Detail).Infoln("Downloading", d.queue.hashPool.Size(), "blocks")
    atomic.StoreInt32(&d.downloadingBlocks, 1)

    start := time.Now()

    // default ticker for re-fetching blocks everynow and then
    ticker := time.NewTicker(20 * time.Millisecond)
out:
    for {
        select {
        case blockPack := <-d.blockCh:
            d.queue.deliver(blockPack.peerId, blockPack.blocks)
            d.peers.setState(blockPack.peerId, idleState)
        case <-ticker.C:
            // If there are unrequested hashes left start fetching
            // from the available peers.
            if d.queue.hashPool.Size() > 0 {
                availablePeers := d.peers.get(idleState)
                if len(availablePeers) == 0 {
                    glog.V(logger.Detail).Infoln("No peers available out of", len(d.peers))
                }

                for _, peer := range availablePeers {
                    // Get a possible chunk. If nil is returned no chunk
                    // could be returned due to no hashes available.
                    chunk := d.queue.get(peer, maxBlockFetch)
                    if chunk == nil {
                        continue
                    }

                    //fmt.Println("fetching for", peer.id)
                    // XXX make fetch blocking.
                    // Fetch the chunk and check for error. If the peer was somehow
                    // already fetching a chunk due to a bug, it will be returned to
                    // the queue
                    if err := peer.fetch(chunk); err != nil {
                        // log for tracing
                        glog.V(logger.Debug).Infof("peer %s received double work (state = %v)\n", peer.id, peer.state)
                        d.queue.put(chunk.hashes)
                    }
                }
                atomic.StoreInt32(&d.downloadingBlocks, 1)
            } else if len(d.queue.fetching) == 0 {
                // When there are no more queue and no more `fetching`. We can
                // safely assume we're done. Another part of the process will  check
                // for parent errors and will re-request anything that's missing
                atomic.StoreInt32(&d.downloadingBlocks, 0)
                // Break out so that we can process with processing blocks
                break out
            } else {
                // Check for bad peers. Bad peers may indicate a peer not responding
                // to a `getBlocks` message. A timeout of 5 seconds is set. Peers
                // that badly or poorly behave are removed from the peer set (not banned).
                // Bad peers are excluded from the available peer set and therefor won't be
                // reused. XXX We could re-introduce peers after X time.
                d.queue.mu.Lock()
                var badPeers []string
                for pid, chunk := range d.queue.fetching {
                    if time.Since(chunk.itime) > 5*time.Second {
                        badPeers = append(badPeers, pid)
                        // remove peer as good peer from peer list
                        d.UnregisterPeer(pid)
                    }
                }
                d.queue.mu.Unlock()

                for _, pid := range badPeers {
                    // A nil chunk is delivered so that the chunk's hashes are given
                    // back to the queue objects. When hashes are put back in the queue
                    // other (decent) peers can pick them up.
                    // XXX We could make use of a reputation system here ranking peers
                    // in their performance
                    // 1) Time for them to respond;
                    // 2) Measure their speed;
                    // 3) Amount and availability.
                    d.queue.deliver(pid, nil)
                }

            }
            //fmt.Println(d.queue.hashPool.Size(), len(d.queue.fetching))
        }
    }

    glog.V(logger.Detail).Infoln("Download blocks: done. Took", time.Since(start))

    return nil
}

// Add an (unrequested) block to the downloader. This is usually done through the
// NewBlockMsg by the protocol handler.
func (d *Downloader) AddBlock(id string, block *types.Block, td *big.Int) {
    hash := block.Hash()

    if d.hasBlock(hash) {
        return
    }

    peer := d.peers.getPeer(id)
    // if the peer is in our healthy list of peers; update the td
    // and add the block. Otherwise just ignore it
    if peer == nil {
        glog.V(logger.Detail).Infof("Ignored block from bad peer %s\n", id)
        return
    }

    peer.mu.Lock()
    peer.td = td
    peer.recentHash = block.Hash()
    peer.mu.Unlock()

    glog.V(logger.Detail).Infoln("Inserting new block from:", id)
    d.queue.addBlock(id, block, td)

    // if neither go ahead to process
    if !(d.isFetchingHashes() || d.isDownloadingBlocks()) {
        // Check if the parent of the received block is known.
        // If the block is not know, request it otherwise, request.
        phash := block.ParentHash()
        if !d.hasBlock(phash) {
            glog.V(logger.Detail).Infof("Missing parent %x, requires fetching\n", phash.Bytes()[:4])
            d.syncCh <- syncPack{peer, peer.recentHash, true}
        } else {
            d.process()
        }
    }
}

// Deliver a chunk to the downloader. This is usually done through the BlocksMsg by
// the protocol handler.
func (d *Downloader) DeliverChunk(id string, blocks []*types.Block) {
    d.blockCh <- blockPack{id, blocks}
}

func (d *Downloader) process() error {
    atomic.StoreInt32(&d.processingBlocks, 1)
    defer atomic.StoreInt32(&d.processingBlocks, 0)

    // XXX this will move when optimised
    // Sort the blocks by number. This bit needs much improvement. Right now
    // it assumes full honesty form peers (i.e. it's not checked when the blocks
    // link). We should at least check whihc queue match. This code could move
    // to a seperate goroutine where it periodically checks for linked pieces.
    types.BlockBy(types.Number).Sort(d.queue.blocks)
    blocks := d.queue.blocks

    glog.V(logger.Debug).Infoln("Inserting chain with", len(blocks), "blocks")

    var err error
    // Loop untill we're out of blocks
    for len(blocks) != 0 {
        max := int(math.Min(float64(len(blocks)), 256))
        // TODO check for parent error. When there's a parent error we should stop
        // processing and start requesting the `block.hash` so that it's parent and
        // grandparents can be requested and queued.
        err = d.insertChain(blocks[:max])
        if err != nil && core.IsParentErr(err) {
            glog.V(logger.Debug).Infoln("Aborting process due to missing parent. Fetching hashes")

            // TODO change this. This shite
            for i, block := range blocks[:max] {
                if !d.hasBlock(block.ParentHash()) {
                    d.syncCh <- syncPack{d.peers.bestPeer(), block.Hash(), true}
                    // remove processed blocks
                    blocks = blocks[i:]

                    break
                }
            }
            break
        }
        blocks = blocks[max:]
    }

    // This will allow the GC to remove the in memory blocks
    if len(blocks) == 0 {
        d.queue.blocks = nil
    } else {
        d.queue.blocks = blocks
    }
    return err
}

func (d *Downloader) isFetchingHashes() bool {
    return atomic.LoadInt32(&d.fetchingHashes) == 1
}

func (d *Downloader) isDownloadingBlocks() bool {
    return atomic.LoadInt32(&d.downloadingBlocks) == 1
}

func (d *Downloader) isProcessing() bool {
    return atomic.LoadInt32(&d.processingBlocks) == 1
}