mirror/lnd
1
0 Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

routing: global probability based mission control

Browse Source 
Previously every payment had its own local mission control state which
was in effect only for that payment. In this commit most of the local
state is removed and payments all tap into the global mission control
probability estimator.

Furthermore the decay time of pruned edges and nodes is extended, so
that observations about the network can better benefit future payment
processes.

Last, the probability function is transformed from a binary output to a
gradual curve, allowing for a better trade off between candidate routes.
This commit is contained in:
Joost Jager committed 2019-06-04 10:00:25 +02:00
1 parent 3349d517aa
commit 7133f37bb8
12 files changed
+338 -220
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

routing/missioncontrol.go
+175 -119
View File
@@ -1,6 +1,7 @@
package routing
import (
"math"
"sync"
"time"
@@ -13,48 +14,27 @@ import (
)
const (
// vertexDecay is the decay period of colored vertexes added to
// MissionControl. Once vertexDecay passes after an entry has been
// added to the prune view, it is garbage collected. This value is
// larger than edgeDecay as an edge failure typical indicates an
// unbalanced channel, while a vertex failure indicates a node is not
// online and active.
vertexDecay = time.Duration(time.Minute * 5)
// defaultPenaltyHalfLife is the default half-life duration. The
// half-life duration defines after how much time a penalized node or
// channel is back at 50% probability.
defaultPenaltyHalfLife = time.Hour
// edgeDecay is the decay period of colored edges added to
// MissionControl. Once edgeDecay passed after an entry has been added,
// it is garbage collected. This value is smaller than vertexDecay as
// an edge related failure during payment sending typically indicates
// that a channel was unbalanced, a condition which may quickly change.
//
// TODO(roasbeef): instead use random delay on each?
edgeDecay = time.Duration(time.Second * 5)
// aprioriHopProbability is the assumed success probability of a hop in
// a route when no other information is available.
aprioriHopProbability = 1
)
// MissionControl contains state which summarizes the past attempts of HTLC
// routing by external callers when sending payments throughout the network.
// MissionControl remembers the outcome of these past routing attempts (success
// and failure), and is able to provide hints/guidance to future HTLC routing
// attempts. MissionControl maintains a decaying network view of the
// edges/vertexes that should be marked as "pruned" during path finding. This
// graph view acts as a shared memory during HTLC payment routing attempts.
// With each execution, if an error is encountered, based on the type of error
// and the location of the error within the route, an edge or vertex is added
// to the view. Later sending attempts will then query the view for all the
// vertexes/edges that should be ignored. Items in the view decay after a set
// period of time, allowing the view to be dynamic w.r.t network changes.
// routing by external callers when sending payments throughout the network. It
// acts as a shared memory during routing attempts with the goal to optimize the
// payment attempt success rate.
//
// Failed payment attempts are reported to mission control. These reports are
// used to track the time of the last node or channel level failure. The time
// since the last failure is used to estimate a success probability that is fed
// into the path finding process for subsequent payment attempts.
type MissionControl struct {
// failedEdges maps a short channel ID to be pruned, to the time that
// it was added to the prune view. Edges are added to this map if a
// caller reports to MissionControl a failure localized to that edge
// when sending a payment.
failedEdges map[EdgeLocator]time.Time
// failedVertexes maps a node's public key that should be pruned, to
// the time that it was added to the prune view. Vertexes are added to
// this map if a caller reports to MissionControl a failure localized
// to that particular vertex.
failedVertexes map[route.Vertex]time.Time
history map[route.Vertex]*nodeHistory
graph *channeldb.ChannelGraph
@@ -62,6 +42,14 @@ type MissionControl struct {
queryBandwidth func(*channeldb.ChannelEdgeInfo) lnwire.MilliSatoshi
// now is expected to return the current time. It is supplied as an
// external function to enable deterministic unit tests.
now func() time.Time
// penaltyHalfLife defines after how much time a penalized node or
// channel is back at 50% probability.
penaltyHalfLife time.Duration
sync.Mutex
// TODO(roasbeef): further counters, if vertex continually unavailable,
@@ -74,83 +62,41 @@ type MissionControl struct {
// PaymentSessionSource interface.
var _ PaymentSessionSource = (*MissionControl)(nil)
// NewMissionControl returns a new instance of MissionControl.
// nodeHistory contains a summary of payment attempt outcomes involving a
// particular node.
type nodeHistory struct {
// lastFail is the last time a node level failure occurred, if any.
lastFail *time.Time
// channelLastFail tracks history per channel, if available for that
// channel.
channelLastFail map[uint64]*channelHistory
}
// channelHistory contains a summary of payment attempt outcomes involving a
// particular channel.
type channelHistory struct {
// lastFail is the last time a channel level failure occurred.
lastFail time.Time
// minPenalizeAmt is the minimum amount for which to take this failure
// into account.
minPenalizeAmt lnwire.MilliSatoshi
}
// NewMissionControl returns a new instance of missionControl.
//
// TODO(roasbeef): persist memory
func NewMissionControl(g *channeldb.ChannelGraph, selfNode *channeldb.LightningNode,
qb func(*channeldb.ChannelEdgeInfo) lnwire.MilliSatoshi) *MissionControl {
return &MissionControl{
failedEdges: make(map[EdgeLocator]time.Time),
failedVertexes: make(map[route.Vertex]time.Time),
selfNode: selfNode,
queryBandwidth: qb,
graph: g,
}
}
// graphPruneView is a filter of sorts that path finding routines should
// consult during the execution. Any edges or vertexes within the view should
// be ignored during path finding. The contents of the view reflect the current
// state of the wider network from the PoV of mission control compiled via HTLC
// routing attempts in the past.
type graphPruneView struct {
edges map[EdgeLocator]struct{}
vertexes map[route.Vertex]struct{}
}
// graphPruneView returns a new graphPruneView instance which is to be
// consulted during path finding. If a vertex/edge is found within the returned
// prune view, it is to be ignored as a goroutine has had issues routing
// through it successfully. Within this method the main view of the
// MissionControl is garbage collected as entries are detected to be "stale".
func (m *MissionControl) graphPruneView() graphPruneView {
// First, we'll grab the current time, this value will be used to
// determine if an entry is stale or not.
now := time.Now()
m.Lock()
// For each of the vertexes that have been added to the prune view, if
// it is now "stale", then we'll ignore it and avoid adding it to the
// view we'll return.
vertexes := make(map[route.Vertex]struct{})
for vertex, pruneTime := range m.failedVertexes {
if now.Sub(pruneTime) >= vertexDecay {
log.Tracef("Pruning decayed failure report for vertex %v "+
"from Mission Control", vertex)
delete(m.failedVertexes, vertex)
continue
}
vertexes[vertex] = struct{}{}
}
// We'll also do the same for edges, but use the edgeDecay this time
// rather than the decay for vertexes.
edges := make(map[EdgeLocator]struct{})
for edge, pruneTime := range m.failedEdges {
if now.Sub(pruneTime) >= edgeDecay {
log.Tracef("Pruning decayed failure report for edge %v "+
"from Mission Control", edge)
delete(m.failedEdges, edge)
continue
}
edges[edge] = struct{}{}
}
m.Unlock()
log.Debugf("Mission Control returning prune view of %v edges, %v "+
"vertexes", len(edges), len(vertexes))
return graphPruneView{
edges: edges,
vertexes: vertexes,
history: make(map[route.Vertex]*nodeHistory),
selfNode: selfNode,
queryBandwidth: qb,
graph: g,
now: time.Now,
penaltyHalfLife: defaultPenaltyHalfLife,
}
}
@@ -161,8 +107,6 @@ func (m *MissionControl) graphPruneView() graphPruneView {
func (m *MissionControl) NewPaymentSession(routeHints [][]zpay32.HopHint,
target route.Vertex) (PaymentSession, error) {
viewSnapshot := m.graphPruneView()
edges := make(map[route.Vertex][]*channeldb.ChannelEdgePolicy)
// Traverse through all of the available hop hints and include them in
@@ -226,10 +170,9 @@ func (m *MissionControl) NewPaymentSession(routeHints [][]zpay32.HopHint,
}
return &paymentSession{
pruneViewSnapshot: viewSnapshot,
additionalEdges: edges,
bandwidthHints: bandwidthHints,
errFailedPolicyChans: make(map[EdgeLocator]struct{}),
errFailedPolicyChans: make(map[nodeChannel]struct{}),
mc: m,
pathFinder: findPath,
}, nil
@@ -239,8 +182,7 @@ func (m *MissionControl) NewPaymentSession(routeHints [][]zpay32.HopHint,
// used for failure reporting to missioncontrol.
func (m *MissionControl) NewPaymentSessionForRoute(preBuiltRoute *route.Route) PaymentSession {
return &paymentSession{
pruneViewSnapshot: m.graphPruneView(),
errFailedPolicyChans: make(map[EdgeLocator]struct{}),
errFailedPolicyChans: make(map[nodeChannel]struct{}),
mc: m,
preBuiltRoute: preBuiltRoute,
}
@@ -251,8 +193,7 @@ func (m *MissionControl) NewPaymentSessionForRoute(preBuiltRoute *route.Route) P
// missioncontrol for resumed payment we don't want to make more attempts for.
func (m *MissionControl) NewPaymentSessionEmpty() PaymentSession {
return &paymentSession{
pruneViewSnapshot: m.graphPruneView(),
errFailedPolicyChans: make(map[EdgeLocator]struct{}),
errFailedPolicyChans: make(map[nodeChannel]struct{}),
mc: m,
preBuiltRoute: &route.Route{},
preBuiltRouteTried: true,
@@ -298,7 +239,122 @@ func generateBandwidthHints(sourceNode *channeldb.LightningNode,
// if no payment attempts have been made.
func (m *MissionControl) ResetHistory() {
m.Lock()
m.failedEdges = make(map[EdgeLocator]time.Time)
m.failedVertexes = make(map[route.Vertex]time.Time)
m.Unlock()
defer m.Unlock()
m.history = make(map[route.Vertex]*nodeHistory)
log.Debugf("Mission control history cleared")
}
// getEdgeProbability is expected to return the success probability of a payment
// from fromNode along edge.
func (m *MissionControl) getEdgeProbability(fromNode route.Vertex,
edge EdgeLocator, amt lnwire.MilliSatoshi) float64 {
m.Lock()
defer m.Unlock()
// Get the history for this node. If there is no history available,
// assume that it's success probability is a constant a priori
// probability. After the attempt new information becomes available to
// adjust this probability.
nodeHistory, ok := m.history[fromNode]
if !ok {
return aprioriHopProbability
}
return m.getEdgeProbabilityForNode(nodeHistory, edge.ChannelID, amt)
}
// getEdgeProbabilityForNode estimates the probability of successfully
// traversing a channel based on the node history.
func (m *MissionControl) getEdgeProbabilityForNode(nodeHistory *nodeHistory,
channelID uint64, amt lnwire.MilliSatoshi) float64 {
// Calculate the last failure of the given edge. A node failure is
// considered a failure that would have affected every edge. Therefore
// we insert a node level failure into the history of every channel.
lastFailure := nodeHistory.lastFail
// Take into account a minimum penalize amount. For balance errors, a
// failure may be reported with such a minimum to prevent too aggresive
// penalization. We only take into account a previous failure if the
// amount that we currently get the probability for is greater or equal
// than the minPenalizeAmt of the previous failure.
channelHistory, ok := nodeHistory.channelLastFail[channelID]
if ok && channelHistory.minPenalizeAmt <= amt {
// If there is both a node level failure recorded and a channel
// level failure is applicable too, we take the most recent of
// the two.
if lastFailure == nil ||
channelHistory.lastFail.After(*lastFailure) {
lastFailure = &channelHistory.lastFail
}
}
if lastFailure == nil {
return aprioriHopProbability
}
timeSinceLastFailure := m.now().Sub(*lastFailure)
// Calculate success probability. It is an exponential curve that brings
// the probability down to zero when a failure occurs. From there it
// recovers asymptotically back to the a priori probability. The rate at
// which this happens is controlled by the penaltyHalfLife parameter.
exp := -timeSinceLastFailure.Hours() / m.penaltyHalfLife.Hours()
probability := aprioriHopProbability * (1 - math.Pow(2, exp))
return probability
}
// createHistoryIfNotExists returns the history for the given node. If the node
// is yet unknown, it will create an empty history structure.
func (m *MissionControl) createHistoryIfNotExists(vertex route.Vertex) *nodeHistory {
if node, ok := m.history[vertex]; ok {
return node
}
node := &nodeHistory{
channelLastFail: make(map[uint64]*channelHistory),
}
m.history[vertex] = node
return node
}
// reportVertexFailure reports a node level failure.
func (m *MissionControl) reportVertexFailure(v route.Vertex) {
log.Debugf("Reporting vertex %v failure to Mission Control", v)
now := m.now()
m.Lock()
defer m.Unlock()
history := m.createHistoryIfNotExists(v)
history.lastFail = &now
}
// reportEdgeFailure reports a channel level failure.
//
// TODO(roasbeef): also add value attempted to send and capacity of channel
func (m *MissionControl) reportEdgeFailure(failedEdge edge,
minPenalizeAmt lnwire.MilliSatoshi) {
log.Debugf("Reporting channel %v failure to Mission Control",
failedEdge.channel)
now := m.now()
m.Lock()
defer m.Unlock()
history := m.createHistoryIfNotExists(failedEdge.from)
history.channelLastFail[failedEdge.channel] = &channelHistory{
lastFail: now,
minPenalizeAmt: minPenalizeAmt,
}
}