akovalenko/lnd.xprv
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge pull request #2916 from cfromknecht/split-syncer-query-reply

Browse Source 
discovery: make gossip replies synchronous
This commit is contained in:
Olaoluwa Osuntokun 2019-04-29 17:40:13 -07:00 committed by GitHub
commit 985902be27
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 350 additions and 564 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
discovery/sync_manager.go
View File

@ -386,6 +386,9 @@ func (m *SyncManager) createGossipSyncer(peer lnpeer.Peer) *GossipSyncer {
sendToPeer: func(msgs ...lnwire.Message) error {
return peer.SendMessageLazy(false, msgs...)
},
sendToPeerSync: func(msgs ...lnwire.Message) error {
return peer.SendMessageLazy(true, msgs...)
},
})
// Gossip syncers are initialized by default in a PassiveSync type

166
discovery/syncer.go
View File

@ -9,6 +9,7 @@ import (
"time"
"github.com/btcsuite/btcd/chaincfg/chainhash"
"github.com/lightningnetwork/lnd/lnpeer"
"github.com/lightningnetwork/lnd/lnwire"
"golang.org/x/time/rate"
)
@ -206,11 +207,16 @@ type gossipSyncerCfg struct {
// the remote node in a single QueryShortChanIDs request.
batchSize int32
// sendToPeer is a function closure that should send the set of
// targeted messages to the peer we've been assigned to sync the graph
// state from.
// sendToPeer sends a variadic number of messages to the remote peer.
// This method should not block while waiting for sends to be written
// to the wire.
sendToPeer func(...lnwire.Message) error
// sendToPeerSync sends a variadic number of messages to the remote
// peer, blocking until all messages have been sent successfully or a
// write error is encountered.
sendToPeerSync func(...lnwire.Message) error
// maxUndelayedQueryReplies specifies how many gossip queries we will
// respond to immediately before starting to delay responses.
maxUndelayedQueryReplies int
@ -219,6 +225,16 @@ type gossipSyncerCfg struct {
// responding to gossip queries after replying to
// maxUndelayedQueryReplies queries.
delayedQueryReplyInterval time.Duration
// noSyncChannels will prevent the GossipSyncer from spawning a
// channelGraphSyncer, meaning we will not try to reconcile unknown
// channels with the remote peer.
noSyncChannels bool
// noReplyQueries will prevent the GossipSyncer from spawning a
// replyHandler, meaning we will not reply to queries from our remote
// peer.
noReplyQueries bool
}
// GossipSyncer is a struct that handles synchronizing the channel graph state
@ -271,10 +287,15 @@ type GossipSyncer struct {
// PassiveSync to ActiveSync.
genHistoricalChanRangeQuery bool
// gossipMsgs is a channel that all messages from the target peer will
// be sent over.
// gossipMsgs is a channel that all responses to our queries from the
// target peer will be sent over, these will be read by the
// channelGraphSyncer.
gossipMsgs chan lnwire.Message
// queryMsgs is a channel that all queries from the remote peer will be
// received over, these will be read by the replyHandler.
queryMsgs chan lnwire.Message
// bufferedChanRangeReplies is used in the waitingQueryChanReply to
// buffer all the chunked response to our query.
bufferedChanRangeReplies []lnwire.ShortChannelID
@ -332,6 +353,7 @@ func newGossipSyncer(cfg gossipSyncerCfg) *GossipSyncer {
syncTransitionReqs: make(chan *syncTransitionReq),
historicalSyncReqs: make(chan *historicalSyncReq),
gossipMsgs: make(chan lnwire.Message, 100),
queryMsgs: make(chan lnwire.Message, 100),
quit: make(chan struct{}),
}
}
@ -342,8 +364,17 @@ func (g *GossipSyncer) Start() {
g.started.Do(func() {
log.Debugf("Starting GossipSyncer(%x)", g.cfg.peerPub[:])
g.wg.Add(1)
go g.channelGraphSyncer()
// TODO(conner): only spawn channelGraphSyncer if remote
// supports gossip queries, and only spawn replyHandler if we
// advertise support
if !g.cfg.noSyncChannels {
g.wg.Add(1)
go g.channelGraphSyncer()
}
if !g.cfg.noReplyQueries {
g.wg.Add(1)
go g.replyHandler()
}
})
}
@ -369,7 +400,7 @@ func (g *GossipSyncer) channelGraphSyncer() {
log.Debugf("GossipSyncer(%x): state=%v, type=%v",
g.cfg.peerPub[:], state, syncType)
switch syncerState(state) {
switch state {
// When we're in this state, we're trying to synchronize our
// view of the network with the remote peer. We'll kick off
// this sync by asking them for the set of channels they
@ -382,14 +413,14 @@ func (g *GossipSyncer) channelGraphSyncer() {
g.genHistoricalChanRangeQuery,
)
if err != nil {
log.Errorf("unable to gen chan range "+
log.Errorf("Unable to gen chan range "+
"query: %v", err)
return
}
err = g.cfg.sendToPeer(queryRangeMsg)
if err != nil {
log.Errorf("unable to send chan range "+
log.Errorf("Unable to send chan range "+
"query: %v", err)
return
}
@ -417,22 +448,16 @@ func (g *GossipSyncer) channelGraphSyncer() {
if ok {
err := g.processChanRangeReply(queryReply)
if err != nil {
log.Errorf("unable to "+
log.Errorf("Unable to "+
"process chan range "+
"query: %v", err)
return
}
continue
}
// Otherwise, it's the remote peer performing a
// query, which we'll attempt to reply to.
err := g.replyPeerQueries(msg)
if err != nil && err != ErrGossipSyncerExiting {
log.Errorf("unable to reply to peer "+
"query: %v", err)
}
log.Warnf("Unexpected message: %T in state=%v",
msg, state)
case <-g.quit:
return
@ -447,7 +472,7 @@ func (g *GossipSyncer) channelGraphSyncer() {
// query chunk.
done, err := g.synchronizeChanIDs()
if err != nil {
log.Errorf("unable to sync chan IDs: %v", err)
log.Errorf("Unable to sync chan IDs: %v", err)
}
// If this wasn't our last query, then we'll need to
@ -480,13 +505,8 @@ func (g *GossipSyncer) channelGraphSyncer() {
continue
}
// Otherwise, it's the remote peer performing a
// query, which we'll attempt to deploy to.
err := g.replyPeerQueries(msg)
if err != nil && err != ErrGossipSyncerExiting {
log.Errorf("unable to reply to peer "+
"query: %v", err)
}
log.Warnf("Unexpected message: %T in state=%v",
msg, state)
case <-g.quit:
return
@ -520,13 +540,6 @@ func (g *GossipSyncer) channelGraphSyncer() {
// messages or process any state transitions and exit if
// needed.
select {
case msg := <-g.gossipMsgs:
err := g.replyPeerQueries(msg)
if err != nil && err != ErrGossipSyncerExiting {
log.Errorf("unable to reply to peer "+
"query: %v", err)
}
case req := <-g.syncTransitionReqs:
req.errChan <- g.handleSyncTransition(req)
@ -540,6 +553,38 @@ func (g *GossipSyncer) channelGraphSyncer() {
}
}
// replyHandler is an event loop whose sole purpose is to reply to the remote
// peers queries. Our replyHandler will respond to messages generated by their
// channelGraphSyncer, and vice versa. Each party's channelGraphSyncer drives
// the other's replyHandler, allowing the replyHandler to operate independently
// from the state machine maintained on the same node.
//
// NOTE: This method MUST be run as a goroutine.
func (g *GossipSyncer) replyHandler() {
defer g.wg.Done()
for {
select {
case msg := <-g.queryMsgs:
err := g.replyPeerQueries(msg)
switch {
case err == ErrGossipSyncerExiting:
return
case err == lnpeer.ErrPeerExiting:
return
case err != nil:
log.Errorf("Unable to reply to peer "+
"query: %v", err)
}
case <-g.quit:
return
}
}
}
// sendGossipTimestampRange constructs and sets a GossipTimestampRange for the
// syncer and sends it to the remote peer.
func (g *GossipSyncer) sendGossipTimestampRange(firstTimestamp time.Time,
@ -818,7 +863,7 @@ func (g *GossipSyncer) replyChanRangeQuery(query *lnwire.QueryChannelRange) erro
if isFinalChunk {
replyChunk.Complete = 1
}
if err := g.cfg.sendToPeer(&replyChunk); err != nil {
if err := g.cfg.sendToPeerSync(&replyChunk); err != nil {
return err
}
@ -846,7 +891,7 @@ func (g *GossipSyncer) replyShortChanIDs(query *lnwire.QueryShortChanIDs) error
"chain=%v, we're on chain=%v", g.cfg.chainHash,
query.ChainHash)
return g.cfg.sendToPeer(&lnwire.ReplyShortChanIDsEnd{
return g.cfg.sendToPeerSync(&lnwire.ReplyShortChanIDsEnd{
ChainHash: query.ChainHash,
Complete: 0,
})
@ -873,23 +918,22 @@ func (g *GossipSyncer) replyShortChanIDs(query *lnwire.QueryShortChanIDs) error
query.ShortChanIDs[0].ToUint64(), err)
}
// If we didn't find any messages related to those channel ID's, then
// we'll send over a reply marking the end of our response, and exit
// early.
if len(replyMsgs) == 0 {
return g.cfg.sendToPeer(&lnwire.ReplyShortChanIDsEnd{
ChainHash: query.ChainHash,
Complete: 1,
})
// Reply with any messages related to those channel ID's, we'll write
// each one individually and synchronously to throttle the sends and
// perform buffering of responses in the syncer as opposed to the peer.
for _, msg := range replyMsgs {
err := g.cfg.sendToPeerSync(msg)
if err != nil {
return err
}
}
// Otherwise, we'll send over our set of messages responding to the
// query, with the ending message appended to it.
replyMsgs = append(replyMsgs, &lnwire.ReplyShortChanIDsEnd{
// Regardless of whether we had any messages to reply with, send over
// the sentinel message to signal that the stream has terminated.
return g.cfg.sendToPeerSync(&lnwire.ReplyShortChanIDsEnd{
ChainHash: query.ChainHash,
Complete: 1,
})
return g.cfg.sendToPeer(replyMsgs...)
}
// ApplyGossipFilter applies a gossiper filter sent by the remote node to the
@ -930,9 +974,19 @@ func (g *GossipSyncer) ApplyGossipFilter(filter *lnwire.GossipTimestampRange) er
go func() {
defer g.wg.Done()
if err := g.cfg.sendToPeer(newUpdatestoSend...); err != nil {
log.Errorf("unable to send messages for peer catch "+
"up: %v", err)
for _, msg := range newUpdatestoSend {
err := g.cfg.sendToPeerSync(msg)
switch {
case err == ErrGossipSyncerExiting:
return
case err == lnpeer.ErrPeerExiting:
return
case err != nil:
log.Errorf("Unable to send message for "+
"peer catch up: %v", err)
}
}
}()
@ -1065,8 +1119,16 @@ func (g *GossipSyncer) FilterGossipMsgs(msgs ...msgWithSenders) {
// ProcessQueryMsg is used by outside callers to pass new channel time series
// queries to the internal processing goroutine.
func (g *GossipSyncer) ProcessQueryMsg(msg lnwire.Message, peerQuit <-chan struct{}) {
var msgChan chan lnwire.Message
switch msg.(type) {
case *lnwire.QueryChannelRange, *lnwire.QueryShortChanIDs:
msgChan = g.queryMsgs
default:
msgChan = g.gossipMsgs
}
select {
case g.gossipMsgs <- msg:
case msgChan <- msg:
case <-peerQuit:
case <-g.quit:
}

722
discovery/syncer_test.go
View File

@ -116,20 +116,45 @@ func (m *mockChannelGraphTimeSeries) FetchChanUpdates(chain chainhash.Hash,
var _ ChannelGraphTimeSeries = (*mockChannelGraphTimeSeries)(nil)
// newTestSyncer creates a new test instance of a GossipSyncer. A buffered
// message channel is returned for intercepting messages sent from the syncer,
// in addition to a mock channel series which allows the test to control which
// messages the syncer knows of or wishes to filter out. The variadic flags are
// treated as positional arguments where the first index signals that the syncer
// should spawn a channelGraphSyncer and second index signals that the syncer
// should spawn a replyHandler. Any flags beyond the first two are currently
// ignored. If no flags are provided, both a channelGraphSyncer and replyHandler
// will be spawned by default.
func newTestSyncer(hID lnwire.ShortChannelID,
encodingType lnwire.ShortChanIDEncoding, chunkSize int32,
) (chan []lnwire.Message, *GossipSyncer, *mockChannelGraphTimeSeries) {
flags ...bool) (chan []lnwire.Message,
*GossipSyncer, *mockChannelGraphTimeSeries) {
syncChannels := true
replyQueries := true
if len(flags) > 0 {
syncChannels = flags[0]
}
if len(flags) > 1 {
replyQueries = flags[1]
}
msgChan := make(chan []lnwire.Message, 20)
cfg := gossipSyncerCfg{
channelSeries: newMockChannelGraphTimeSeries(hID),
encodingType: encodingType,
chunkSize: chunkSize,
batchSize: chunkSize,
channelSeries: newMockChannelGraphTimeSeries(hID),
encodingType: encodingType,
chunkSize: chunkSize,
batchSize: chunkSize,
noSyncChannels: !syncChannels,
noReplyQueries: !replyQueries,
sendToPeer: func(msgs ...lnwire.Message) error {
msgChan <- msgs
return nil
},
sendToPeerSync: func(msgs ...lnwire.Message) error {
msgChan <- msgs
return nil
},
delayedQueryReplyInterval: 2 * time.Second,
}
syncer := newGossipSyncer(cfg)
@ -519,30 +544,37 @@ func TestGossipSyncerReplyShortChanIDs(t *testing.T) {
t.Fatalf("unable to query for chan IDs: %v", err)
}
select {
case <-time.After(time.Second * 15):
t.Fatalf("no msgs received")
for i := 0; i < len(queryReply)+1; i++ {
select {
case <-time.After(time.Second * 15):
t.Fatalf("no msgs received")
// We should get back exactly 4 messages. The first 3 are the same
// messages we sent above, and the query end message.
case msgs := <-msgChan:
if len(msgs) != 4 {
t.Fatalf("wrong messages: expected %v, got %v",
4, len(msgs))
}
// We should get back exactly 4 messages. The first 3 are the
// same messages we sent above, and the query end message.
case msgs := <-msgChan:
if len(msgs) != 1 {
t.Fatalf("wrong number of messages: "+
"expected %v, got %v", 1, len(msgs))
}
if !reflect.DeepEqual(queryReply, msgs[:3]) {
t.Fatalf("wrong set of messages: expected %v, got %v",
spew.Sdump(queryReply), spew.Sdump(msgs[:3]))
}
isQueryReply := i < len(queryReply)
finalMsg, ok := msgs[0].(*lnwire.ReplyShortChanIDsEnd)
finalMsg, ok := msgs[3].(*lnwire.ReplyShortChanIDsEnd)
if !ok {
t.Fatalf("expected lnwire.ReplyShortChanIDsEnd "+
"instead got %T", msgs[3])
}
if finalMsg.Complete != 1 {
t.Fatalf("complete wasn't set")
switch {
case isQueryReply &&
!reflect.DeepEqual(queryReply[i], msgs[0]):
t.Fatalf("wrong message: expected %v, got %v",
spew.Sdump(queryReply[i]),
spew.Sdump(msgs[0]))
case !isQueryReply && !ok:
t.Fatalf("expected lnwire.ReplyShortChanIDsEnd"+
" instead got %T", msgs[3])
case !isQueryReply && finalMsg.Complete != 1:
t.Fatalf("complete wasn't set")
}
}
}
}
@ -1020,13 +1052,13 @@ func TestGossipSyncerDelayDOS(t *testing.T) {
BlockHeight: 1144,
}
msgChan1, syncer1, chanSeries1 := newTestSyncer(
startHeight, defaultEncoding, chunkSize,
startHeight, defaultEncoding, chunkSize, true, false,
)
syncer1.Start()
defer syncer1.Stop()
msgChan2, syncer2, chanSeries2 := newTestSyncer(
startHeight, defaultEncoding, chunkSize,
startHeight, defaultEncoding, chunkSize, false, true,
)
syncer2.Start()
defer syncer2.Stop()
@ -1038,34 +1070,31 @@ func TestGossipSyncerDelayDOS(t *testing.T) {
numUndelayedQueries := syncer1.cfg.maxUndelayedQueryReplies
// We will send enough queries to exhaust the undelayed responses, and
// then send two more queries which should be delayed.
numQueryResponses := numUndelayedQueries + numDelayedQueries
// then send two more queries which should be delayed. An additional one
// is subtracted from the total since undelayed message will be consumed
// by the initial QueryChannelRange.
numQueryResponses := numUndelayedQueries + numDelayedQueries - 1
// The total number of responses must include the initial reply each
// syner will make to QueryChannelRange.
// syncer will make to QueryChannelRange.
numTotalQueries := 1 + numQueryResponses
// The total number of channels each syncer needs to request must be
// scaled by the chunk size being used.
numTotalChans := numQueryResponses * chunkSize
// Although both nodes are at the same height, they'll have a
// completely disjoint set of chan ID's that they know of.
var syncer1Chans []lnwire.ShortChannelID
for i := 0; i < numTotalChans; i++ {
syncer1Chans = append(
syncer1Chans, lnwire.NewShortChanIDFromInt(uint64(i)),
)
}
// Construct enough channels so that all of the queries will have enough
// channels. Since syncer1 won't know of any channels, their sets are
// inherently disjoint.
var syncer2Chans []lnwire.ShortChannelID
for i := numTotalChans; i < numTotalChans+numTotalChans; i++ {
for i := 0; i < numTotalChans; i++ {
syncer2Chans = append(
syncer2Chans, lnwire.NewShortChanIDFromInt(uint64(i)),
)
}
// We'll kick off the test by passing over the QueryChannelRange
// messages from one node to the other.
// We'll kick off the test by asserting syncer1 sends over the
// QueryChannelRange message the other node.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
@ -1083,46 +1112,16 @@ func TestGossipSyncerDelayDOS(t *testing.T) {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer2")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a QueryChannelRange message.
_, ok := msg.(*lnwire.QueryChannelRange)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
case syncer2.queryMsgs <- msg:
}
}
}
// At this point, we'll need to send responses to both nodes from their
// respective channel series. Both nodes will simply request the entire
// set of channels from the other. This will count as the first
// undelayed response for each syncer.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
case <-chanSeries1.filterRangeReqs:
// We'll send all the channels that it should know of.
chanSeries1.filterRangeResp <- syncer1Chans
}
// At this point, we'll need to a response from syncer2's channel
// series. This will cause syncer1 to simply request the entire set of
// channels from syncer2. This will count as the first undelayed
// response for sycner2.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
@ -1132,31 +1131,9 @@ func TestGossipSyncerDelayDOS(t *testing.T) {
chanSeries2.filterRangeResp <- syncer2Chans
}
// At this point, we'll forward the ReplyChannelRange messages to both
// parties. After receiving the set of channels known to the remote peer
// At this point, we'll assert that the ReplyChannelRange message is
// sent by sycner2.
for i := 0; i < numQueryResponses; i++ {
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a ReplyChannelRange message.
_, ok := msg.(*lnwire.ReplyChannelRange)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer2")
@ -1180,8 +1157,7 @@ func TestGossipSyncerDelayDOS(t *testing.T) {
}
}
// We'll now send back a chunked response for both parties of the known
// short chan ID's.
// We'll now have syncer1 process the received sids from syncer2.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
@ -1189,207 +1165,140 @@ func TestGossipSyncerDelayDOS(t *testing.T) {
case <-chanSeries1.filterReq:
chanSeries1.filterResp <- syncer2Chans
}
// At this point, syncer1 should start to send out initial requests to
// query the chan IDs of the remote party. We'll keep track of the
// number of queries made using the iterated value, which starts at one
// due the initial contribution of the QueryChannelRange msgs.
for i := 1; i < numTotalQueries; i++ {
expDelayResponse := i >= numUndelayedQueries
queryBatch(t,
msgChan1, msgChan2,
syncer1, syncer2,
chanSeries2,
expDelayResponse,
delayedQueryInterval,
delayTolerance,
)
}
}
// queryBatch is a helper method that will query for a single batch of channels
// from a peer and assert the responses. The method can also be used to assert
// the same transition happens, but is delayed by the remote peer's DOS
// rate-limiting. The provided chanSeries should belong to syncer2.
//
// The state transition performed is the following:
// syncer1 -- QueryShortChanIDs --> syncer2
// chanSeries.FetchChanAnns()
// syncer1 <-- ReplyShortChanIDsEnd -- syncer2
//
// If expDelayResponse is true, this method will assert that the call the
// FetchChanAnns happens between:
// [delayedQueryInterval-delayTolerance, delayedQueryInterval+delayTolerance].
func queryBatch(t *testing.T,
msgChan1, msgChan2 chan []lnwire.Message,
syncer1, syncer2 *GossipSyncer,
chanSeries *mockChannelGraphTimeSeries,
expDelayResponse bool,
delayedQueryInterval, delayTolerance time.Duration) {
t.Helper()
// First, we'll assert that syncer1 sends a QueryShortChanIDs message to
// the remote peer.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
t.Fatalf("didn't get msg from syncer2")
case <-chanSeries2.filterReq:
chanSeries2.filterResp <- syncer1Chans
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a QueryShortChanIDs message.
_, ok := msg.(*lnwire.QueryShortChanIDs)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryShortChanIDs for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.queryMsgs <- msg:
}
}
}
// At this point, both parties should start to send out initial
// requests to query the chan IDs of the remote party. We'll keep track
// of the number of queries made using the iterated value, which starts
// at one due the initial contribution of the QueryChannelRange msgs.
for i := 1; i < numTotalQueries; i++ {
// Both parties should now have sent out the initial requests
// to query the chan IDs of the other party.
// We'll then respond to with an empty set of replies (as it doesn't
// affect the test).
switch {
// If this query has surpassed the undelayed query threshold, we will
// impose stricter timing constraints on the response times. We'll first
// test that syncer2's chanSeries doesn't immediately receive a query,
// and then check that the query hasn't gone unanswered entirely.
case expDelayResponse:
// Create a before and after timeout to test, our test
// will ensure the messages are delivered to the peer
// in this timeframe.
before := time.After(
delayedQueryInterval - delayTolerance,
)
after := time.After(
delayedQueryInterval + delayTolerance,
)
// First, ensure syncer2 doesn't try to respond up until the
// before time fires.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case <-before:
// Query is delayed, proceed.
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a QueryShortChanIDs message.
_, ok := msg.(*lnwire.QueryShortChanIDs)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryShortChanIDs for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
case <-chanSeries.annReq:
t.Fatalf("DOSy query was not delayed")
}
// If syncer2 doesn't attempt a response within the allowed
// interval, then the messages are probably lost.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer2")
case <-after:
t.Fatalf("no delayed query received")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a QueryShortChanIDs message.
_, ok := msg.(*lnwire.QueryShortChanIDs)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryShortChanIDs for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
}
}
case <-chanSeries.annReq:
chanSeries.annResp <- []lnwire.Message{}
}
// We'll then respond to both parties with an empty set of
// replies (as it doesn't affect the test).
switch {
// If this query has surpassed the undelayed query threshold, we
// will impose stricter timing constraints on the response
// times. We'll first test that the peers don't immediately
// receive a query, and then check that both queries haven't
// gone unanswered entirely.
case i >= numUndelayedQueries:
// Create a before and after timeout to test, our test
// will ensure the messages are delivered to the peers
// in this timeframe.
before := time.After(
delayedQueryInterval - delayTolerance,
)
after := time.After(
delayedQueryInterval + delayTolerance,
)
// First, ensure neither peer tries to respond up until
// the before time fires.
select {
case <-before:
// Queries are delayed, proceed.
case <-chanSeries1.annReq:
t.Fatalf("DOSy query was not delayed")
case <-chanSeries2.annReq:
t.Fatalf("DOSy query was not delayed")
}
// Next, we'll need to test that both queries are
// received before the after timer expires. To account
// for ordering, we will try to pull a message from both
// peers, and then test that the opposite peer also
// receives the message promptly.
var (
firstChanSeries *mockChannelGraphTimeSeries
laterChanSeries *mockChannelGraphTimeSeries
)
// If neither peer attempts a response within the
// allowed interval, then the messages are probably
// lost. Otherwise, process the message and record the
// induced ordering.
select {
case <-after:
t.Fatalf("no delayed query received")
case <-chanSeries1.annReq:
chanSeries1.annResp <- []lnwire.Message{}
firstChanSeries = chanSeries1
laterChanSeries = chanSeries2
case <-chanSeries2.annReq:
chanSeries2.annResp <- []lnwire.Message{}
firstChanSeries = chanSeries2
laterChanSeries = chanSeries1
}
// Finally, using the same interval timeout as before,
// ensure the later peer also responds promptly. We also
// assert that the first peer doesn't attempt another
// response.
select {
case <-after:
t.Fatalf("no delayed query received")
case <-firstChanSeries.annReq:
t.Fatalf("spurious undelayed response")
case <-laterChanSeries.annReq:
laterChanSeries.annResp <- []lnwire.Message{}
}
// Otherwise, we still haven't exceeded our undelayed query
// limit. Assert that both peers promptly attempt a response to
// the queries.
default:
select {
case <-time.After(50 * time.Millisecond):
t.Fatalf("no query recvd")
case <-chanSeries1.annReq:
chanSeries1.annResp <- []lnwire.Message{}
}
select {
case <-time.After(50 * time.Millisecond):
t.Fatalf("no query recvd")
case <-chanSeries2.annReq:
chanSeries2.annResp <- []lnwire.Message{}
}
}
// Finally, both sides should then receive a
// ReplyShortChanIDsEnd as the first chunk has been replied to.
// Otherwise, syncer2 should query its chanSeries promtly.
default:
select {
case <-time.After(50 * time.Millisecond):
t.Fatalf("didn't get msg from syncer1")
t.Fatalf("no query recvd")
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a ReplyShortChanIDsEnd message.
_, ok := msg.(*lnwire.ReplyShortChanIDsEnd)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
case <-chanSeries.annReq:
chanSeries.annResp <- []lnwire.Message{}
}
select {
case <-time.After(50 * time.Millisecond):
t.Fatalf("didn't get msg from syncer2")
}
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a ReplyShortChanIDsEnd message.
_, ok := msg.(*lnwire.ReplyShortChanIDsEnd)
if !ok {
t.Fatalf("wrong message: expected "+
"ReplyShortChanIDsEnd for %T", msg)
}
// Finally, assert that syncer2 replies to syncer1 with a
// ReplyShortChanIDsEnd.
select {
case <-time.After(50 * time.Millisecond):
t.Fatalf("didn't get msg from syncer2")
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a ReplyShortChanIDsEnd message.
_, ok := msg.(*lnwire.ReplyShortChanIDsEnd)
if !ok {
t.Fatalf("wrong message: expected "+
"ReplyShortChanIDsEnd for %T", msg)
}
case syncer1.gossipMsgs <- msg:
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
}
case syncer1.gossipMsgs <- msg:
}
}
}
@ -1413,24 +1322,19 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
BlockHeight: 1144,
}
msgChan1, syncer1, chanSeries1 := newTestSyncer(
startHeight, defaultEncoding, chunkSize,
startHeight, defaultEncoding, chunkSize, true, false,
)
syncer1.Start()
defer syncer1.Stop()
msgChan2, syncer2, chanSeries2 := newTestSyncer(
startHeight, defaultEncoding, chunkSize,
startHeight, defaultEncoding, chunkSize, false, true,
)
syncer2.Start()
defer syncer2.Stop()
// Although both nodes are at the same height, they'll have a
// completely disjoint set of 3 chan ID's that they know of.
syncer1Chans := []lnwire.ShortChannelID{
lnwire.NewShortChanIDFromInt(1),
lnwire.NewShortChanIDFromInt(2),
lnwire.NewShortChanIDFromInt(3),
}
// Although both nodes are at the same height, syncer will have 3 chan
// ID's that syncer1 doesn't know of.
syncer2Chans := []lnwire.ShortChannelID{
lnwire.NewShortChanIDFromInt(4),
lnwire.NewShortChanIDFromInt(5),
@ -1438,7 +1342,7 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
}
// We'll kick off the test by passing over the QueryChannelRange
// messages from one node to the other.
// messages from syncer1 to syncer2.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
@ -1456,45 +1360,15 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer2")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a QueryChannelRange message.
_, ok := msg.(*lnwire.QueryChannelRange)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
case syncer2.queryMsgs <- msg:
}
}
}
// At this point, we'll need to send responses to both nodes from their
// respective channel series. Both nodes will simply request the entire
// set of channels from the other.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
case <-chanSeries1.filterRangeReqs:
// We'll send all the channels that it should know of.
chanSeries1.filterRangeResp <- syncer1Chans
}
// At this point, we'll need to send a response from syncer2 to syncer1
// using syncer2's channels This will cause syncer1 to simply request
// the entire set of channels from the other.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
@ -1504,32 +1378,9 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
chanSeries2.filterRangeResp <- syncer2Chans
}
// At this point, we'll forward the ReplyChannelRange messages to both
// parties. Two replies are expected since the chunk size is 2, and we
// need to query for 3 channels.
for i := 0; i < chunkSize; i++ {
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a ReplyChannelRange message.
_, ok := msg.(*lnwire.ReplyChannelRange)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
}
}
// At this point, we'll assert that syncer2 replies with the
// ReplyChannelRange messages. Two replies are expected since the chunk
// size is 2, and we need to query for 3 channels.
for i := 0; i < chunkSize; i++ {
select {
case <-time.After(time.Second * 2):
@ -1554,8 +1405,7 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
}
}
// We'll now send back a chunked response for both parties of the known
// short chan ID's.
// We'll now send back a chunked response from syncer2 back to sycner1.
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
@ -1563,133 +1413,21 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
case <-chanSeries1.filterReq:
chanSeries1.filterResp <- syncer2Chans
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
case <-chanSeries2.filterReq:
chanSeries2.filterResp <- syncer1Chans
}
// At this point, both parties should start to send out initial
// requests to query the chan IDs of the remote party. As the chunk
// size is 2, they'll need 2 rounds in order to fully reconcile the
// state.
// At this point, syncer1 should start to send out initial requests to
// query the chan IDs of the remote party. As the chunk size is 2,
// they'll need 2 rounds in order to fully reconcile the state.
for i := 0; i < chunkSize; i++ {
// Both parties should now have sent out the initial requests
// to query the chan IDs of the other party.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a QueryShortChanIDs message.
_, ok := msg.(*lnwire.QueryShortChanIDs)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryShortChanIDs for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer2")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a QueryShortChanIDs message.
_, ok := msg.(*lnwire.QueryShortChanIDs)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryShortChanIDs for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
}
}
}
// We'll then respond to both parties with an empty set of replies (as
// it doesn't affect the test).
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
case <-chanSeries1.annReq:
chanSeries1.annResp <- []lnwire.Message{}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("no query recvd")
case <-chanSeries2.annReq:
chanSeries2.annResp <- []lnwire.Message{}
}
// Both sides should then receive a ReplyShortChanIDsEnd as the first
// chunk has been replied to.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case msgs := <-msgChan1:
for _, msg := range msgs {
// The message MUST be a ReplyShortChanIDsEnd message.
_, ok := msg.(*lnwire.ReplyShortChanIDsEnd)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
}
}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a ReplyShortChanIDsEnd message.
_, ok := msg.(*lnwire.ReplyShortChanIDsEnd)
if !ok {
t.Fatalf("wrong message: expected "+
"ReplyShortChanIDsEnd for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
}
}
}
queryBatch(t,
msgChan1, msgChan2,
syncer1, syncer2,
chanSeries2,
false, 0, 0,
)
}
// At this stage both parties should now be sending over their initial
// GossipTimestampRange messages as they should both be fully synced.
// At this stage syncer1 should now be sending over its initial
// GossipTimestampRange messages as it should be fully synced.
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
@ -1709,28 +1447,6 @@ func TestGossipSyncerRoutineSync(t *testing.T) {
case syncer2.gossipMsgs <- msg:
}
}
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("didn't get msg from syncer1")
case msgs := <-msgChan2:
for _, msg := range msgs {
// The message MUST be a GossipTimestampRange message.
_, ok := msg.(*lnwire.GossipTimestampRange)
if !ok {
t.Fatalf("wrong message: expected "+
"QueryChannelRange for %T", msg)
}
select {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
}
}
}
@ -1796,7 +1512,7 @@ func TestGossipSyncerAlreadySynced(t *testing.T) {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer2.gossipMsgs <- msg:
case syncer2.queryMsgs <- msg:
}
}
@ -1818,7 +1534,7 @@ func TestGossipSyncerAlreadySynced(t *testing.T) {
case <-time.After(time.Second * 2):
t.Fatalf("node 2 didn't read msg")
case syncer1.gossipMsgs <- msg:
case syncer1.queryMsgs <- msg:
}
}

8
lnpeer/errors.go Normal file
View File

@ -0,0 +1,8 @@
package lnpeer
import "fmt"
var (
// ErrPeerExiting signals that the peer received a disconnect request.
ErrPeerExiting = fmt.Errorf("peer exiting")
)

15
peer.go
View File

@ -32,9 +32,6 @@ import (
var (
numNodes int32
// ErrPeerExiting signals that the peer received a disconnect request.
ErrPeerExiting = fmt.Errorf("peer exiting")
)
const (
@ -1411,7 +1408,7 @@ func (p *peer) logWireMessage(msg lnwire.Message, read bool) {
func (p *peer) writeMessage(msg lnwire.Message) error {
// Simply exit if we're shutting down.
if atomic.LoadInt32(&p.disconnect) != 0 {
return ErrPeerExiting
return lnpeer.ErrPeerExiting
}
// Only log the message on the first attempt.
@ -1559,7 +1556,7 @@ out:
}
case <-p.quit:
exitErr = ErrPeerExiting
exitErr = lnpeer.ErrPeerExiting
break out
}
}
@ -1691,7 +1688,7 @@ func (p *peer) queue(priority bool, msg lnwire.Message, errChan chan error) {
case <-p.quit:
peerLog.Tracef("Peer shutting down, could not enqueue msg.")
if errChan != nil {
errChan <- ErrPeerExiting
errChan <- lnpeer.ErrPeerExiting
}
}
}
@ -2504,7 +2501,7 @@ func (p *peer) sendMessage(sync, priority bool, msgs ...lnwire.Message) error {
case err := <-errChan:
return err
case <-p.quit:
return ErrPeerExiting
return lnpeer.ErrPeerExiting
}
}
@ -2550,7 +2547,7 @@ func (p *peer) AddNewChannel(channel *channeldb.OpenChannel,
case <-cancel:
return errors.New("canceled adding new channel")
case <-p.quit:
return ErrPeerExiting
return lnpeer.ErrPeerExiting
}
// We pause here to wait for the peer to recognize the new channel
@ -2559,7 +2556,7 @@ func (p *peer) AddNewChannel(channel *channeldb.OpenChannel,
case err := <-errChan:
return err
case <-p.quit:
return ErrPeerExiting
return lnpeer.ErrPeerExiting
}
}