providerquerymanager.go 10.1 KB
Newer Older
1 2 3 4 5
package providerquerymanager

import (
	"context"
	"sync"
6
	"time"
7 8 9 10 11 12 13 14 15 16 17

	"github.com/ipfs/go-cid"
	logging "github.com/ipfs/go-log"
	peer "github.com/libp2p/go-libp2p-peer"
)

var log = logging.Logger("bitswap")

const (
	maxProviders         = 10
	maxInProcessRequests = 6
18
	defaultTimeout       = 10 * time.Second
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
)

type inProgressRequestStatus struct {
	providersSoFar []peer.ID
	listeners      map[uint64]chan peer.ID
}

// ProviderQueryNetwork is an interface for finding providers and connecting to
// peers.
type ProviderQueryNetwork interface {
	ConnectTo(context.Context, peer.ID) error
	FindProvidersAsync(context.Context, cid.Cid, int) <-chan peer.ID
}

type providerQueryMessage interface {
	handle(pqm *ProviderQueryManager)
}

type receivedProviderMessage struct {
	k cid.Cid
	p peer.ID
}

type finishedProviderQueryMessage struct {
	k cid.Cid
}

type newProvideQueryMessage struct {
	ses                   uint64
	k                     cid.Cid
	inProgressRequestChan chan<- inProgressRequest
}

type cancelRequestMessage struct {
	ses uint64
	k   cid.Cid
}

// ProviderQueryManager manages requests to find more providers for blocks
// for bitswap sessions. It's main goals are to:
// - rate limit requests -- don't have too many find provider calls running
// simultaneously
// - connect to found peers and filter them if it can't connect
// - ensure two findprovider calls for the same block don't run concurrently
// - manage timeouts
type ProviderQueryManager struct {
65 66 67
	ctx                          context.Context
	network                      ProviderQueryNetwork
	providerQueryMessages        chan providerQueryMessage
68 69
	providerRequestsProcessing   chan cid.Cid
	incomingFindProviderRequests chan cid.Cid
70 71 72 73 74 75

	findProviderTimeout time.Duration
	timeoutMutex        sync.RWMutex

	// do not touch outside the run loop
	inProgressRequestStatuses map[cid.Cid]*inProgressRequestStatus
76 77 78 79 80 81 82 83 84 85 86 87
}

// New initializes a new ProviderQueryManager for a given context and a given
// network provider.
func New(ctx context.Context, network ProviderQueryNetwork) *ProviderQueryManager {
	return &ProviderQueryManager{
		ctx:                          ctx,
		network:                      network,
		providerQueryMessages:        make(chan providerQueryMessage, 16),
		providerRequestsProcessing:   make(chan cid.Cid),
		incomingFindProviderRequests: make(chan cid.Cid),
		inProgressRequestStatuses:    make(map[cid.Cid]*inProgressRequestStatus),
88
		findProviderTimeout:          defaultTimeout,
89 90 91 92 93 94 95 96 97 98 99 100 101
	}
}

// Startup starts processing for the ProviderQueryManager.
func (pqm *ProviderQueryManager) Startup() {
	go pqm.run()
}

type inProgressRequest struct {
	providersSoFar []peer.ID
	incoming       <-chan peer.ID
}

102 103 104 105 106 107 108
// SetFindProviderTimeout changes the timeout for finding providers
func (pqm *ProviderQueryManager) SetFindProviderTimeout(findProviderTimeout time.Duration) {
	pqm.timeoutMutex.Lock()
	pqm.findProviderTimeout = findProviderTimeout
	pqm.timeoutMutex.Unlock()
}

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
// FindProvidersAsync finds providers for the given block.
func (pqm *ProviderQueryManager) FindProvidersAsync(sessionCtx context.Context, k cid.Cid, ses uint64) <-chan peer.ID {
	inProgressRequestChan := make(chan inProgressRequest)

	select {
	case pqm.providerQueryMessages <- &newProvideQueryMessage{
		ses:                   ses,
		k:                     k,
		inProgressRequestChan: inProgressRequestChan,
	}:
	case <-pqm.ctx.Done():
		return nil
	case <-sessionCtx.Done():
		return nil
	}

	var receivedInProgressRequest inProgressRequest
	select {
	case <-sessionCtx.Done():
		return nil
	case receivedInProgressRequest = <-inProgressRequestChan:
	}

	return pqm.receiveProviders(sessionCtx, k, ses, receivedInProgressRequest)
}

func (pqm *ProviderQueryManager) receiveProviders(sessionCtx context.Context, k cid.Cid, ses uint64, receivedInProgressRequest inProgressRequest) <-chan peer.ID {
	// maintains an unbuffered queue for incoming providers for given request for a given session
	// essentially, as a provider comes in, for a given CID, we want to immediately broadcast to all
	// sessions that queried that CID, without worrying about whether the client code is actually
	// reading from the returned channel -- so that the broadcast never blocks
	// based on: https://medium.com/capital-one-tech/building-an-unbounded-channel-in-go-789e175cd2cd
	returnedProviders := make(chan peer.ID)
	receivedProviders := append([]peer.ID(nil), receivedInProgressRequest.providersSoFar[0:]...)
	incomingProviders := receivedInProgressRequest.incoming

	go func() {
		defer close(returnedProviders)
		outgoingProviders := func() chan<- peer.ID {
			if len(receivedProviders) == 0 {
				return nil
			}
			return returnedProviders
		}
		nextProvider := func() peer.ID {
			if len(receivedProviders) == 0 {
				return ""
			}
			return receivedProviders[0]
		}
		for len(receivedProviders) > 0 || incomingProviders != nil {
			select {
			case <-sessionCtx.Done():
				pqm.providerQueryMessages <- &cancelRequestMessage{
					ses: ses,
					k:   k,
				}
				// clear out any remaining providers
				for range incomingProviders {
				}
				return
			case provider, ok := <-incomingProviders:
				if !ok {
					incomingProviders = nil
				} else {
					receivedProviders = append(receivedProviders, provider)
				}
			case outgoingProviders() <- nextProvider():
				receivedProviders = receivedProviders[1:]
			}
		}
	}()
	return returnedProviders
}

func (pqm *ProviderQueryManager) findProviderWorker() {
	// findProviderWorker just cycles through incoming provider queries one
	// at a time. We have six of these workers running at once
	// to let requests go in parallel but keep them rate limited
	for {
		select {
		case k, ok := <-pqm.providerRequestsProcessing:
			if !ok {
				return
			}

195 196 197 198 199
			pqm.timeoutMutex.RLock()
			findProviderCtx, cancel := context.WithTimeout(pqm.ctx, pqm.findProviderTimeout)
			pqm.timeoutMutex.RUnlock()
			defer cancel()
			providers := pqm.network.FindProvidersAsync(findProviderCtx, k, maxProviders)
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
			wg := &sync.WaitGroup{}
			for p := range providers {
				wg.Add(1)
				go func(p peer.ID) {
					defer wg.Done()
					err := pqm.network.ConnectTo(pqm.ctx, p)
					if err != nil {
						log.Debugf("failed to connect to provider %s: %s", p, err)
						return
					}
					select {
					case pqm.providerQueryMessages <- &receivedProviderMessage{
						k: k,
						p: p,
					}:
					case <-pqm.ctx.Done():
						return
					}
				}(p)
			}
			wg.Wait()
			select {
			case pqm.providerQueryMessages <- &finishedProviderQueryMessage{
				k: k,
			}:
			case <-pqm.ctx.Done():
			}
		case <-pqm.ctx.Done():
			return
		}
	}
}

func (pqm *ProviderQueryManager) providerRequestBufferWorker() {
	// the provider request buffer worker just maintains an unbounded
	// buffer for incoming provider queries and dispatches to the find
	// provider workers as they become available
	// based on: https://medium.com/capital-one-tech/building-an-unbounded-channel-in-go-789e175cd2cd
	var providerQueryRequestBuffer []cid.Cid
	nextProviderQuery := func() cid.Cid {
		if len(providerQueryRequestBuffer) == 0 {
			return cid.Cid{}
		}
		return providerQueryRequestBuffer[0]
	}
	outgoingRequests := func() chan<- cid.Cid {
		if len(providerQueryRequestBuffer) == 0 {
			return nil
		}
		return pqm.providerRequestsProcessing
	}

	for {
		select {
		case incomingRequest, ok := <-pqm.incomingFindProviderRequests:
			if !ok {
				return
			}
			providerQueryRequestBuffer = append(providerQueryRequestBuffer, incomingRequest)
		case outgoingRequests() <- nextProviderQuery():
			providerQueryRequestBuffer = providerQueryRequestBuffer[1:]
		case <-pqm.ctx.Done():
			return
		}
	}
}

func (pqm *ProviderQueryManager) cleanupInProcessRequests() {
	for _, requestStatus := range pqm.inProgressRequestStatuses {
		for _, listener := range requestStatus.listeners {
			close(listener)
		}
	}
}

func (pqm *ProviderQueryManager) run() {
	defer close(pqm.incomingFindProviderRequests)
	defer close(pqm.providerRequestsProcessing)
	defer pqm.cleanupInProcessRequests()

	go pqm.providerRequestBufferWorker()
	for i := 0; i < maxInProcessRequests; i++ {
		go pqm.findProviderWorker()
	}

	for {
		select {
		case nextMessage := <-pqm.providerQueryMessages:
			nextMessage.handle(pqm)
		case <-pqm.ctx.Done():
			return
		}
	}
}

func (rpm *receivedProviderMessage) handle(pqm *ProviderQueryManager) {
	requestStatus, ok := pqm.inProgressRequestStatuses[rpm.k]
	if !ok {
		log.Errorf("Received provider (%s) for cid (%s) not requested", rpm.p.String(), rpm.k.String())
		return
	}
	requestStatus.providersSoFar = append(requestStatus.providersSoFar, rpm.p)
	for _, listener := range requestStatus.listeners {
		select {
		case listener <- rpm.p:
		case <-pqm.ctx.Done():
			return
		}
	}
}

func (fpqm *finishedProviderQueryMessage) handle(pqm *ProviderQueryManager) {
	requestStatus, ok := pqm.inProgressRequestStatuses[fpqm.k]
	if !ok {
		log.Errorf("Ended request for cid (%s) not in progress", fpqm.k.String())
		return
	}
	for _, listener := range requestStatus.listeners {
		close(listener)
	}
	delete(pqm.inProgressRequestStatuses, fpqm.k)
}

func (npqm *newProvideQueryMessage) handle(pqm *ProviderQueryManager) {
	requestStatus, ok := pqm.inProgressRequestStatuses[npqm.k]
	if !ok {
		requestStatus = &inProgressRequestStatus{
			listeners: make(map[uint64]chan peer.ID),
		}
		pqm.inProgressRequestStatuses[npqm.k] = requestStatus
		select {
		case pqm.incomingFindProviderRequests <- npqm.k:
		case <-pqm.ctx.Done():
			return
		}
	}
	requestStatus.listeners[npqm.ses] = make(chan peer.ID)
	select {
	case npqm.inProgressRequestChan <- inProgressRequest{
		providersSoFar: requestStatus.providersSoFar,
		incoming:       requestStatus.listeners[npqm.ses],
	}:
	case <-pqm.ctx.Done():
	}
}

func (crm *cancelRequestMessage) handle(pqm *ProviderQueryManager) {
	requestStatus, ok := pqm.inProgressRequestStatuses[crm.k]
	if !ok {
		log.Errorf("Attempt to cancel request for session (%d) for cid (%s) not in progress", crm.ses, crm.k.String())
		return
	}
	listener, ok := requestStatus.listeners[crm.ses]
	if !ok {
		log.Errorf("Attempt to cancel request for session (%d) for cid (%s) this is not a listener", crm.ses, crm.k.String())
		return
	}
	close(listener)
	delete(requestStatus.listeners, crm.ses)
}