diff options
Diffstat (limited to 'eth/downloader/downloader.go')
-rw-r--r-- | eth/downloader/downloader.go | 148 |
1 files changed, 123 insertions, 25 deletions
diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go index 4e795af6d..c71cfa684 100644 --- a/eth/downloader/downloader.go +++ b/eth/downloader/downloader.go @@ -1,6 +1,7 @@ package downloader import ( + "errors" "math" "math/big" "sync" @@ -16,8 +17,15 @@ import ( ) const ( - maxBlockFetch = 256 // Amount of max blocks to be fetched per chunk - minDesiredPeerCount = 3 // Amount of peers desired to start syncing + maxBlockFetch = 256 // Amount of max blocks to be fetched per chunk + minDesiredPeerCount = 3 // Amount of peers desired to start syncing + blockTtl = 15 * time.Second // The amount of time it takes for a request to time out +) + +var ( + errLowTd = errors.New("peer's TD is too low") + errBusy = errors.New("busy") + errUnknownPeer = errors.New("peer's unknown or unhealthy") ) type hashCheckFn func(common.Hash) bool @@ -82,7 +90,7 @@ func (d *Downloader) RegisterPeer(id string, td *big.Int, hash common.Hash, getH d.mu.Lock() defer d.mu.Unlock() - glog.V(logger.Detail).Infoln("Register peer", id) + glog.V(logger.Detail).Infoln("Register peer", id, "TD =", td) // Create a new peer and add it to the list of known peers peer := newPeer(id, td, hash, getHashes, getBlocks) @@ -94,6 +102,7 @@ func (d *Downloader) RegisterPeer(id string, td *big.Int, hash common.Hash, getH return nil } +// UnregisterPeer unregister's a peer. This will prevent any action from the specified peer. func (d *Downloader) UnregisterPeer(id string) { d.mu.Lock() defer d.mu.Unlock() @@ -103,6 +112,73 @@ func (d *Downloader) UnregisterPeer(id string) { delete(d.peers, id) } +// SynchroniseWithPeer will select the peer and use it for synchronising. If an empty string is given +// it will use the best peer possible and synchronise if it's TD is higher than our own. If any of the +// checks fail an error will be returned. This method is synchronous +func (d *Downloader) SynchroniseWithPeer(id string) (types.Blocks, error) { + // Check if we're busy + if d.isBusy() { + return nil, errBusy + } + + // Attempt to select a peer. This can either be nothing, which returns, best peer + // or selected peer. If no peer could be found an error will be returned + var p *peer + if len(id) == 0 { + p = d.peers[id] + if p == nil { + return nil, errUnknownPeer + } + } else { + p = d.peers.bestPeer() + } + + // Make sure our td is lower than the peer's td + if p.td.Cmp(d.currentTd()) <= 0 || d.hasBlock(p.recentHash) { + return nil, errLowTd + } + + // Get the hash from the peer and initiate the downloading progress. + err := d.getFromPeer(p, p.recentHash, false) + if err != nil { + return nil, err + } + + return d.queue.blocks, nil +} + +// Synchronise will synchronise using the best peer. +func (d *Downloader) Synchronise() (types.Blocks, error) { + return d.SynchroniseWithPeer("") +} + +func (d *Downloader) getFromPeer(p *peer, hash common.Hash, ignoreInitial bool) error { + glog.V(logger.Detail).Infoln("Synchronising with the network using:", p.id) + // Start the fetcher. This will block the update entirely + // interupts need to be send to the appropriate channels + // respectively. + if err := d.startFetchingHashes(p, hash, ignoreInitial); err != nil { + // handle error + glog.V(logger.Debug).Infoln("Error fetching hashes:", err) + // XXX Reset + return err + } + + // 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(p); err != nil { + glog.V(logger.Debug).Infoln("Error downloading blocks:", err) + // XXX reset + return err + } + + glog.V(logger.Detail).Infoln("Sync completed") + + return nil +} + func (d *Downloader) peerHandler() { // itimer is used to determine when to start ignoring `minDesiredPeerCount` //itimer := time.NewTicker(5 * time.Second) @@ -116,11 +192,18 @@ out: 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()) + // Either select the peer or restart the timer if no peers could + // be selected. + if peer := d.peers.bestPeer(); peer != nil { + d.selectPeer(d.peers.bestPeer()) + } else { + itimer.Reset(5 * time.Second) + } case <-d.quit: break out } @@ -131,7 +214,7 @@ 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()) { + if !d.isBusy() { // 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. @@ -142,6 +225,7 @@ func (d *Downloader) selectPeer(p *peer) { glog.V(logger.Detail).Infoln("New peer with highest TD =", p.td) d.syncCh <- syncPack{p, p.recentHash, false} } + } func (d *Downloader) update() { @@ -150,7 +234,7 @@ out: select { case sync := <-d.syncCh: selectedPeer := sync.peer - glog.V(logger.Detail).Infoln("Synchronising with network using:", selectedPeer.id) + 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. @@ -182,6 +266,9 @@ out: // XXX Make synchronous func (d *Downloader) startFetchingHashes(p *peer, hash common.Hash, ignoreInitial bool) error { + atomic.StoreInt32(&d.fetchingHashes, 1) + defer atomic.StoreInt32(&d.fetchingHashes, 0) + glog.V(logger.Debug).Infof("Downloading hashes (%x) from %s", hash.Bytes()[:4], p.id) start := time.Now() @@ -192,10 +279,8 @@ func (d *Downloader) startFetchingHashes(p *peer, hash common.Hash, ignoreInitia // 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 { @@ -216,14 +301,16 @@ out: d.queue.put(hashSet) // Add hashes to the chunk set - // Check if we're done fetching - if !done && len(hashes) > 0 { + if len(hashes) == 0 { // Make sure the peer actually gave you something valid + glog.V(logger.Debug).Infof("Peer (%s) responded with empty hash set\n", p.id) + d.queue.reset() + + break out + } else if !done { // Check if we're done fetching //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) + } else { // we're done break out } } @@ -236,6 +323,7 @@ out: func (d *Downloader) startFetchingBlocks(p *peer) error { glog.V(logger.Detail).Infoln("Downloading", d.queue.hashPool.Size(), "blocks") atomic.StoreInt32(&d.downloadingBlocks, 1) + defer atomic.StoreInt32(&d.downloadingBlocks, 0) start := time.Now() @@ -245,18 +333,18 @@ out: for { select { case blockPack := <-d.blockCh: - d.peers[blockPack.peerId].promote() - d.queue.deliver(blockPack.peerId, blockPack.blocks) - d.peers.setState(blockPack.peerId, idleState) + // If the peer was previously banned and failed to deliver it's pack + // in a reasonable time frame, ignore it's message. + if d.peers[blockPack.peerId] != nil { + d.peers[blockPack.peerId].promote() + 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. @@ -281,8 +369,6 @@ out: // 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 @@ -293,7 +379,7 @@ out: d.queue.mu.Lock() var badPeers []string for pid, chunk := range d.queue.fetching { - if time.Since(chunk.itime) > 5*time.Second { + if time.Since(chunk.itime) > blockTtl { badPeers = append(badPeers, pid) // remove peer as good peer from peer list d.UnregisterPeer(pid) @@ -353,7 +439,7 @@ func (d *Downloader) AddBlock(id string, block *types.Block, td *big.Int) { d.queue.addBlock(id, block, td) // if neither go ahead to process - if !(d.isFetchingHashes() || d.isDownloadingBlocks()) { + if !d.isBusy() { // Check if the parent of the received block is known. // If the block is not know, request it otherwise, request. phash := block.ParentHash() @@ -383,8 +469,11 @@ func (d *Downloader) process() error { // to a seperate goroutine where it periodically checks for linked pieces. types.BlockBy(types.Number).Sort(d.queue.blocks) blocks := d.queue.blocks + if len(blocks) == 0 { + return nil + } - glog.V(logger.Debug).Infoln("Inserting chain with", len(blocks), "blocks") + glog.V(logger.Debug).Infof("Inserting chain with %d blocks (#%v - #%v)\n", len(blocks), blocks[0].Number(), blocks[len(blocks)-1].Number()) var err error // Loop untill we're out of blocks @@ -408,6 +497,11 @@ func (d *Downloader) process() error { } } break + } else if err != nil { + // Reset chain completely. This needs much, much improvement. + // instead: check all blocks leading down to this block false block and remove it + blocks = nil + break } blocks = blocks[max:] } @@ -432,3 +526,7 @@ func (d *Downloader) isDownloadingBlocks() bool { func (d *Downloader) isProcessing() bool { return atomic.LoadInt32(&d.processingBlocks) == 1 } + +func (d *Downloader) isBusy() bool { + return d.isFetchingHashes() || d.isDownloadingBlocks() || d.isProcessing() +} |