forked from ipfs/kubo
-
Notifications
You must be signed in to change notification settings - Fork 1
/
bitswap.go
463 lines (389 loc) · 12.7 KB
/
bitswap.go
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
// package bitswap implements the IPFS exchange interface with the BitSwap
// bilateral exchange protocol.
package bitswap
import (
"context"
"errors"
"math"
"sync"
"sync/atomic"
"time"
blockstore "github.com/ipfs/go-ipfs/blocks/blockstore"
exchange "github.com/ipfs/go-ipfs/exchange"
decision "github.com/ipfs/go-ipfs/exchange/bitswap/decision"
bsmsg "github.com/ipfs/go-ipfs/exchange/bitswap/message"
bsnet "github.com/ipfs/go-ipfs/exchange/bitswap/network"
notifications "github.com/ipfs/go-ipfs/exchange/bitswap/notifications"
flags "github.com/ipfs/go-ipfs/flags"
"github.com/ipfs/go-ipfs/thirdparty/delay"
cid "gx/ipfs/QmNp85zy9RLrQ5oQD4hPyS39ezrrXpcaa7R4Y9kxdWQLLQ/go-cid"
metrics "gx/ipfs/QmRg1gKTHzc3CZXSKzem8aR4E3TubFhbgXwfVuWnSK5CC5/go-metrics-interface"
process "gx/ipfs/QmSF8fPo3jgVBAy8fpdjjYqgG87dkJgUprRBHRd2tmfgpP/goprocess"
procctx "gx/ipfs/QmSF8fPo3jgVBAy8fpdjjYqgG87dkJgUprRBHRd2tmfgpP/goprocess/context"
blocks "gx/ipfs/QmSn9Td7xgxm9EV7iEjTckpUWmWApggzPxu7eFGWkkpwin/go-block-format"
logging "gx/ipfs/QmSpJByNKFX1sCsHBEp3R73FL4NF6FnQTEGyNAXHm2GS52/go-log"
peer "gx/ipfs/QmXYjuNuxVzXKJCfWasQk1RqkhVLDM9jtUKhqc2WPQmFSB/go-libp2p-peer"
)
var log = logging.Logger("bitswap")
const (
// maxProvidersPerRequest specifies the maximum number of providers desired
// from the network. This value is specified because the network streams
// results.
// TODO: if a 'non-nice' strategy is implemented, consider increasing this value
maxProvidersPerRequest = 3
providerRequestTimeout = time.Second * 10
provideTimeout = time.Second * 15
sizeBatchRequestChan = 32
// kMaxPriority is the max priority as defined by the bitswap protocol
kMaxPriority = math.MaxInt32
)
var (
HasBlockBufferSize = 256
provideKeysBufferSize = 2048
provideWorkerMax = 512
// the 1<<18+15 is to observe old file chunks that are 1<<18 + 14 in size
metricsBuckets = []float64{1 << 6, 1 << 10, 1 << 14, 1 << 18, 1<<18 + 15, 1 << 22}
)
func init() {
if flags.LowMemMode {
HasBlockBufferSize = 64
provideKeysBufferSize = 512
provideWorkerMax = 16
}
}
var rebroadcastDelay = delay.Fixed(time.Minute)
// New initializes a BitSwap instance that communicates over the provided
// BitSwapNetwork. This function registers the returned instance as the network
// delegate.
// Runs until context is cancelled.
func New(parent context.Context, p peer.ID, network bsnet.BitSwapNetwork,
bstore blockstore.Blockstore, nice bool) exchange.Interface {
// important to use provided parent context (since it may include important
// loggable data). It's probably not a good idea to allow bitswap to be
// coupled to the concerns of the ipfs daemon in this way.
//
// FIXME(btc) Now that bitswap manages itself using a process, it probably
// shouldn't accept a context anymore. Clients should probably use Close()
// exclusively. We should probably find another way to share logging data
ctx, cancelFunc := context.WithCancel(parent)
ctx = metrics.CtxSubScope(ctx, "bitswap")
dupHist := metrics.NewCtx(ctx, "recv_dup_blocks_bytes", "Summary of duplicate"+
" data blocks recived").Histogram(metricsBuckets)
allHist := metrics.NewCtx(ctx, "recv_all_blocks_bytes", "Summary of all"+
" data blocks recived").Histogram(metricsBuckets)
notif := notifications.New()
px := process.WithTeardown(func() error {
notif.Shutdown()
return nil
})
bs := &Bitswap{
blockstore: bstore,
notifications: notif,
engine: decision.NewEngine(ctx, bstore), // TODO close the engine with Close() method
network: network,
findKeys: make(chan *blockRequest, sizeBatchRequestChan),
process: px,
newBlocks: make(chan *cid.Cid, HasBlockBufferSize),
provideKeys: make(chan *cid.Cid, provideKeysBufferSize),
wm: NewWantManager(ctx, network),
counters: new(counters),
dupMetric: dupHist,
allMetric: allHist,
}
go bs.wm.Run()
network.SetDelegate(bs)
// Start up bitswaps async worker routines
bs.startWorkers(px, ctx)
// bind the context and process.
// do it over here to avoid closing before all setup is done.
go func() {
<-px.Closing() // process closes first
cancelFunc()
}()
procctx.CloseAfterContext(px, ctx) // parent cancelled first
return bs
}
// Bitswap instances implement the bitswap protocol.
type Bitswap struct {
// the peermanager manages sending messages to peers in a way that
// wont block bitswap operation
wm *WantManager
// the engine is the bit of logic that decides who to send which blocks to
engine *decision.Engine
// network delivers messages on behalf of the session
network bsnet.BitSwapNetwork
// blockstore is the local database
// NB: ensure threadsafety
blockstore blockstore.Blockstore
// notifications engine for receiving new blocks and routing them to the
// appropriate user requests
notifications notifications.PubSub
// findKeys sends keys to a worker to find and connect to providers for them
findKeys chan *blockRequest
// newBlocks is a channel for newly added blocks to be provided to the
// network. blocks pushed down this channel get buffered and fed to the
// provideKeys channel later on to avoid too much network activity
newBlocks chan *cid.Cid
// provideKeys directly feeds provide workers
provideKeys chan *cid.Cid
process process.Process
// Counters for various statistics
counterLk sync.Mutex
counters *counters
// Metrics interface metrics
dupMetric metrics.Histogram
allMetric metrics.Histogram
// Sessions
sessions []*Session
sessLk sync.Mutex
sessID uint64
sessIDLk sync.Mutex
}
type counters struct {
blocksRecvd uint64
dupBlocksRecvd uint64
dupDataRecvd uint64
blocksSent uint64
dataSent uint64
dataRecvd uint64
messagesRecvd uint64
}
type blockRequest struct {
Cid *cid.Cid
Ctx context.Context
}
// GetBlock attempts to retrieve a particular block from peers within the
// deadline enforced by the context.
func (bs *Bitswap) GetBlock(parent context.Context, k *cid.Cid) (blocks.Block, error) {
return getBlock(parent, k, bs.GetBlocks)
}
func (bs *Bitswap) WantlistForPeer(p peer.ID) []*cid.Cid {
var out []*cid.Cid
for _, e := range bs.engine.WantlistForPeer(p) {
out = append(out, e.Cid)
}
return out
}
func (bs *Bitswap) LedgerForPeer(p peer.ID) *decision.Receipt {
return bs.engine.LedgerForPeer(p)
}
// GetBlocks returns a channel where the caller may receive blocks that
// correspond to the provided |keys|. Returns an error if BitSwap is unable to
// begin this request within the deadline enforced by the context.
//
// NB: Your request remains open until the context expires. To conserve
// resources, provide a context with a reasonably short deadline (ie. not one
// that lasts throughout the lifetime of the server)
func (bs *Bitswap) GetBlocks(ctx context.Context, keys []*cid.Cid) (<-chan blocks.Block, error) {
if len(keys) == 0 {
out := make(chan blocks.Block)
close(out)
return out, nil
}
select {
case <-bs.process.Closing():
return nil, errors.New("bitswap is closed")
default:
}
promise := bs.notifications.Subscribe(ctx, keys...)
for _, k := range keys {
log.Event(ctx, "Bitswap.GetBlockRequest.Start", k)
}
mses := bs.getNextSessionID()
bs.wm.WantBlocks(ctx, keys, nil, mses)
// NB: Optimization. Assumes that providers of key[0] are likely to
// be able to provide for all keys. This currently holds true in most
// every situation. Later, this assumption may not hold as true.
req := &blockRequest{
Cid: keys[0],
Ctx: ctx,
}
remaining := cid.NewSet()
for _, k := range keys {
remaining.Add(k)
}
out := make(chan blocks.Block)
go func() {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
defer close(out)
defer func() {
// can't just defer this call on its own, arguments are resolved *when* the defer is created
bs.CancelWants(remaining.Keys(), mses)
}()
for {
select {
case blk, ok := <-promise:
if !ok {
return
}
bs.CancelWants([]*cid.Cid{blk.Cid()}, mses)
remaining.Remove(blk.Cid())
select {
case out <- blk:
case <-ctx.Done():
return
}
case <-ctx.Done():
return
}
}
}()
select {
case bs.findKeys <- req:
return out, nil
case <-ctx.Done():
return nil, ctx.Err()
}
}
func (bs *Bitswap) getNextSessionID() uint64 {
bs.sessIDLk.Lock()
defer bs.sessIDLk.Unlock()
bs.sessID++
return bs.sessID
}
// CancelWant removes a given key from the wantlist
func (bs *Bitswap) CancelWants(cids []*cid.Cid, ses uint64) {
if len(cids) == 0 {
return
}
bs.wm.CancelWants(context.Background(), cids, nil, ses)
}
// HasBlock announces the existance of a block to this bitswap service. The
// service will potentially notify its peers.
func (bs *Bitswap) HasBlock(blk blocks.Block) error {
return bs.receiveBlockFrom(blk, "")
}
// TODO: Some of this stuff really only needs to be done when adding a block
// from the user, not when receiving it from the network.
// In case you run `git blame` on this comment, I'll save you some time: ask
// @whyrusleeping, I don't know the answers you seek.
func (bs *Bitswap) receiveBlockFrom(blk blocks.Block, from peer.ID) error {
select {
case <-bs.process.Closing():
return errors.New("bitswap is closed")
default:
}
err := bs.blockstore.Put(blk)
if err != nil {
log.Errorf("Error writing block to datastore: %s", err)
return err
}
// NOTE: There exists the possiblity for a race condition here. If a user
// creates a node, then adds it to the dagservice while another goroutine
// is waiting on a GetBlock for that object, they will receive a reference
// to the same node. We should address this soon, but i'm not going to do
// it now as it requires more thought and isnt causing immediate problems.
bs.notifications.Publish(blk)
k := blk.Cid()
ks := []*cid.Cid{k}
for _, s := range bs.SessionsForBlock(k) {
s.receiveBlockFrom(from, blk)
bs.CancelWants(ks, s.id)
}
bs.engine.AddBlock(blk)
select {
case bs.newBlocks <- blk.Cid():
// send block off to be reprovided
case <-bs.process.Closing():
return bs.process.Close()
}
return nil
}
// SessionsForBlock returns a slice of all sessions that may be interested in the given cid
func (bs *Bitswap) SessionsForBlock(c *cid.Cid) []*Session {
bs.sessLk.Lock()
defer bs.sessLk.Unlock()
var out []*Session
for _, s := range bs.sessions {
if s.interestedIn(c) {
out = append(out, s)
}
}
return out
}
func (bs *Bitswap) ReceiveMessage(ctx context.Context, p peer.ID, incoming bsmsg.BitSwapMessage) {
atomic.AddUint64(&bs.counters.messagesRecvd, 1)
// This call records changes to wantlists, blocks received,
// and number of bytes transfered.
bs.engine.MessageReceived(p, incoming)
// TODO: this is bad, and could be easily abused.
// Should only track *useful* messages in ledger
iblocks := incoming.Blocks()
if len(iblocks) == 0 {
return
}
// quickly send out cancels, reduces chances of duplicate block receives
var keys []*cid.Cid
for _, block := range iblocks {
if _, found := bs.wm.wl.Contains(block.Cid()); !found {
log.Infof("received un-asked-for %s from %s", block, p)
continue
}
keys = append(keys, block.Cid())
}
wg := sync.WaitGroup{}
for _, block := range iblocks {
wg.Add(1)
go func(b blocks.Block) { // TODO: this probably doesnt need to be a goroutine...
defer wg.Done()
bs.updateReceiveCounters(b)
log.Debugf("got block %s from %s", b, p)
if err := bs.receiveBlockFrom(b, p); err != nil {
log.Warningf("ReceiveMessage recvBlockFrom error: %s", err)
}
log.Event(ctx, "Bitswap.GetBlockRequest.End", b.Cid())
}(block)
}
wg.Wait()
}
var ErrAlreadyHaveBlock = errors.New("already have block")
func (bs *Bitswap) updateReceiveCounters(b blocks.Block) {
blkLen := len(b.RawData())
has, err := bs.blockstore.Has(b.Cid())
if err != nil {
log.Infof("blockstore.Has error: %s", err)
return
}
bs.allMetric.Observe(float64(blkLen))
if has {
bs.dupMetric.Observe(float64(blkLen))
}
bs.counterLk.Lock()
defer bs.counterLk.Unlock()
c := bs.counters
c.blocksRecvd++
c.dataRecvd += uint64(len(b.RawData()))
if has {
c.dupBlocksRecvd++
c.dupDataRecvd += uint64(blkLen)
}
}
// Connected/Disconnected warns bitswap about peer connections
func (bs *Bitswap) PeerConnected(p peer.ID) {
bs.wm.Connected(p)
bs.engine.PeerConnected(p)
}
// Connected/Disconnected warns bitswap about peer connections
func (bs *Bitswap) PeerDisconnected(p peer.ID) {
bs.wm.Disconnected(p)
bs.engine.PeerDisconnected(p)
}
func (bs *Bitswap) ReceiveError(err error) {
log.Infof("Bitswap ReceiveError: %s", err)
// TODO log the network error
// TODO bubble the network error up to the parent context/error logger
}
func (bs *Bitswap) Close() error {
return bs.process.Close()
}
func (bs *Bitswap) GetWantlist() []*cid.Cid {
var out []*cid.Cid
for _, e := range bs.wm.wl.Entries() {
out = append(out, e.Cid)
}
return out
}
func (bs *Bitswap) IsOnline() bool {
return true
}