eth: limit number of sent transactions based on message size

Nodes that are out of sync will queue many transactions, which causes
the initial transactions message to grow very large. Larger transactions
messages can make communication impossible if the message is too big to
send. Big transactions messages also exhaust egress bandwidth, which
degrades other peer connections.

The new approach to combat these issues is to send transactions in
smaller batches. This commit introduces a new goroutine that handles
delivery of all initial transaction transfers. Size-limited packs of
transactions are sent to one peer at a time, conserving precious egress
bandwidth.

eth/sync.go

@@ -2,6 +2,7 @@ package eth
 
 import (
 	"math"
+	"math/rand"
 	"sync/atomic"
 	"time"
 
@@ -10,6 +11,7 @@ import (
 	"github.com/ethereum/go-ethereum/eth/downloader"
 	"github.com/ethereum/go-ethereum/logger"
 	"github.com/ethereum/go-ethereum/logger/glog"
+	"github.com/ethereum/go-ethereum/p2p/discover"
 )
 
 const (
@@ -20,6 +22,10 @@ const (
 	notifyFetchTimeout  = 5 * time.Second        // Maximum alloted time to return an explicitly requested block
 	minDesiredPeerCount = 5                      // Amount of peers desired to start syncing
 	blockProcAmount     = 256
+
+	// This is the target size for the packs of transactions sent by txsyncLoop.
+	// A pack can get larger than this if a single transactions exceeds this size.
+	txsyncPackSize = 100 * 1024
 )
 
 // blockAnnounce is the hash notification of the availability of a new block in
@@ -30,6 +36,94 @@ type blockAnnounce struct {
 	time time.Time
 }
 
+type txsync struct {
+	p   *peer
+	txs []*types.Transaction
+}
+
+// syncTransactions starts sending all currently pending transactions to the given peer.
+func (pm *ProtocolManager) syncTransactions(p *peer) {
+	txs := pm.txpool.GetTransactions()
+	if len(txs) == 0 {
+		return
+	}
+	select {
+	case pm.txsyncCh <- &txsync{p, txs}:
+	case <-pm.quitSync:
+	}
+}
+
+// txsyncLoop takes care of the initial transaction sync for each new
+// connection. When a new peer appears, we relay all currently pending
+// transactions. In order to minimise egress bandwidth usage, we send
+// the transactions in small packs to one peer at a time.
+func (pm *ProtocolManager) txsyncLoop() {
+	var (
+		pending = make(map[discover.NodeID]*txsync)
+		sending = false               // whether a send is active
+		pack    = new(txsync)         // the pack that is being sent
+		done    = make(chan error, 1) // result of the send
+	)
+
+	// send starts a sending a pack of transactions from the sync.
+	send := func(s *txsync) {
+		// Fill pack with transactions up to the target size.
+		size := common.StorageSize(0)
+		pack.p = s.p
+		pack.txs = pack.txs[:0]
+		for i := 0; i < len(s.txs) && size < txsyncPackSize; i++ {
+			pack.txs = append(pack.txs, s.txs[i])
+			size += s.txs[i].Size()
+		}
+		// Remove the transactions that will be sent.
+		s.txs = s.txs[:copy(s.txs, s.txs[len(pack.txs):])]
+		if len(s.txs) == 0 {
+			delete(pending, s.p.ID())
+		}
+		// Send the pack in the background.
+		glog.V(logger.Detail).Infof("%v: sending %d transactions (%v)", s.p.Peer, len(pack.txs), size)
+		sending = true
+		go func() { done <- pack.p.sendTransactions(pack.txs) }()
+	}
+
+	// pick chooses the next pending sync.
+	pick := func() *txsync {
+		if len(pending) == 0 {
+			return nil
+		}
+		n := rand.Intn(len(pending)) + 1
+		for _, s := range pending {
+			if n--; n == 0 {
+				return s
+			}
+		}
+		return nil
+	}
+
+	for {
+		select {
+		case s := <-pm.txsyncCh:
+			pending[s.p.ID()] = s
+			if !sending {
+				send(s)
+			}
+		case err := <-done:
+			sending = false
+			// Stop tracking peers that cause send failures.
+			if err != nil {
+				glog.V(logger.Debug).Infof("%v: tx send failed: %v", pack.p.Peer, err)
+				delete(pending, pack.p.ID())
+			}
+			// Schedule the next send.
+			if s := pick(); s != nil {
+				send(s)
+			}
+		case <-pm.quitSync:
+			return
+		}
+	}
+}
+
 // fetcher is responsible for collecting hash notifications, and periodically
 // checking all unknown ones and individually fetching them.
 func (pm *ProtocolManager) fetcher() {