lnd/routing/pathfind.go

346 lines
13 KiB
Go
Raw Normal View History

package routing
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
import (
"math"
"container/heap"
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
"github.com/lightningnetwork/lnd/channeldb"
"github.com/roasbeef/btcd/btcec"
"github.com/roasbeef/btcd/chaincfg/chainhash"
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
"github.com/roasbeef/btcutil"
)
const (
// HopLimit is the maximum number hops that is permissible as a route.
// Any potential paths found that lie above this limit will be rejected
// with an error. This value is computed using the current fixed-size
// packet length of the Sphinx construction.
HopLimit = 20
// infinity is used as a starting distance in our shortest path search.
infinity = math.MaxFloat64
)
// Route represents a path through the channel graph which runs over one or
// more channels in succession. This struct carries all the information
// required to craft the Sphinx onion packet, and send the payment along the
// first hop in the path. A route is only selected as valid if all the channels
// have sufficient capacity to carry the initial payment amount after fees are
// accounted for.
type Route struct {
// TotalTimeLock is the cumulative (final) time lock across the entire
// route. This is the CLTV value that should be extended to the first
// hop in the route. All other hops will decrement the time-lock as
// advertised, leaving enough time for all hops to wait for or present
// the payment preimage to complete the payment.
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
TotalTimeLock uint32
// TotalFees is the sum of the fees paid at each hop within the final
// route. In the case of a one-hop payment, this value will be zero as
// we don't need to pay a fee it ourself.
TotalFees btcutil.Amount
// TotalAmount is the total amount of funds required to complete a
// payment over this route. This value includes the cumulative fees at
// each hop. As a result, the HTLC extended to the first-hop in the
// route will need to have at least this many satoshis, otherwise the
// route will fail at an intermediate node due to an insufficient
// amount of fees.
TotalAmount btcutil.Amount
// Hops contains details concerning the specific forwarding details at
// each hop.
Hops []*Hop
}
// Hop represents the forwarding details at a particular position within the
// final route. This struct houses the values necessary to create the HTLC
// which will travel along this hop, and also encode the per-hop payload
// included within the Sphinx packet.
type Hop struct {
// Channel is the active payment channel edge that this hop will travel
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// along.
Channel *ChannelHop
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// TimeLockDelta is the delta that this hop will subtract from the HTLC
// before extending it to the next hop in the route.
TimeLockDelta uint16
// AmtToForward is the amount that this hop will forward to the next
// hop. This value is less than the value that the incoming HTLC
// carries as a fee will be subtracted by the hop.
AmtToForward btcutil.Amount
// Fee is the total fee that this hop will subtract from the incoming
// payment, this difference nets the hop fees for forwarding the
// payment.
Fee btcutil.Amount
}
// ChannelHop is an intermediate hop within the network with a greater
// multi-hop payment route. This struct contains the relevant routing policy of
// the particular edge, as well as the total capacity, and origin chain of the
// channel itself.
type ChannelHop struct {
// Capacity is the total capacity of the channel being traversed. This
// value is expressed for stability in satoshis.
Capacity btcutil.Amount
// Chain is a 32-byte has that denotes the base blockchain network of
// the channel. The 32-byte hash is the "genesis" block of the
// blockchain, or the very first block in the chain.
//
// TODO(roasbeef): store chain within edge info/policy in database.
Chain chainhash.Hash
*channeldb.ChannelEdgePolicy
}
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// computeFee computes the fee to forward an HTLC of `amt` satoshis over the
// passed active payment channel. This value is currently computed as specified
// in BOLT07, but will likely change in the near future.
func computeFee(amt btcutil.Amount, edge *ChannelHop) btcutil.Amount {
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
return edge.FeeBaseMSat + (amt*edge.FeeProportionalMillionths)/1000000
}
// newRoute returns a fully valid route between the source and target that's
// capable of supporting a payment of `amtToSend` after fees are fully
// computed. If the route is too long, or the selected path cannot support the
// fully payment including fees, then a non-nil error is returned.
//
// NOTE: The passed slice of ChannelHops MUST be sorted in reverse order: from
// the target to the source node of the path finding aattempt.
func newRoute(amtToSend btcutil.Amount, pathEdges []*ChannelHop) (*Route, error) {
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
route := &Route{
Hops: make([]*Hop, len(pathEdges)),
}
// The running amount is the total amount of satoshis required at this
// point in the route. We start this value at the amount we want to
// send to the destination. This value will then get successively
// larger as we compute the fees going backwards.
runningAmt := amtToSend
pathLength := len(pathEdges)
for i, edge := range pathEdges {
// Now we create the hop struct for this point in the route.
// The amount to forward is the running amount, and we compute
// the required fee based on this amount.
nextHop := &Hop{
Channel: edge,
AmtToForward: runningAmt,
Fee: computeFee(runningAmt, edge),
TimeLockDelta: edge.TimeLockDelta,
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
}
edge.Node.PubKey.Curve = nil
// As a sanity check, we ensure that the selected channel has
// enough capacity to forward the required amount which
// includes the fee dictated at each hop.
if nextHop.AmtToForward > nextHop.Channel.Capacity {
return nil, ErrInsufficientCapacity
}
// We don't pay any fees to ourselves on the first-hop channel,
// so we don't tally up the running fee and amount.
if i != len(pathEdges)-1 {
// For a node to forward an HTLC, then following
// inequality most hold true: amt_in - fee >=
// amt_to_forward. Therefore we add the fee this node
// consumes in order to calculate the amount that it
// show be forwarded by the prior node which is the
// next hop in our loop.
runningAmt += nextHop.Fee
// Next we tally the total fees (thus far) in the
// route, and also accumulate the total timelock in the
// route by adding the node's time lock delta which is
// the amount of blocks it'll subtract from the
// incoming time lock.
route.TotalFees += nextHop.Fee
} else {
nextHop.Fee = 0
}
route.TotalTimeLock += uint32(nextHop.TimeLockDelta)
// Finally, as we're currently talking the route backwards, we
// reverse the index in order to place this hop at the proper
// spot in the forward direction of the route.
route.Hops[pathLength-1-i] = nextHop
}
// The total amount required for this route will be the value the
// source extends to the first hop in the route.
route.TotalAmount = runningAmt
return route, nil
}
// vertex is a simple alias for the serialization of a compressed Bitcoin
// public key.
type vertex [33]byte
// newVertex returns a new vertex given a public key.
func newVertex(pub *btcec.PublicKey) vertex {
var v vertex
copy(v[:], pub.SerializeCompressed())
return v
}
// edgeWithPrev is a helper struct used in path finding that couples an
// directional edge with the node's ID in the opposite direction.
type edgeWithPrev struct {
edge *ChannelHop
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
prevNode *btcec.PublicKey
}
// edgeWeight computes the weight of an edge. This value is used when searching
// for the shortest path within the channel graph between two nodes. Currently
// this is just 1 + the cltv delta value required at this hop, this value
// should be tuned with experimental and empirical data.
//
// TODO(roasbeef): compute robust weight metric
func edgeWeight(e *channeldb.ChannelEdgePolicy) float64 {
return float64(1 + e.TimeLockDelta)
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
}
// findRoute attempts to find a path from the source node within the
// ChannelGraph to the target node that's capable of supporting a payment of
// `amt` value. The current approach implemented is modified version of
// Dijkstra's algorithm to find a single shortest path between the source node
// and the destination. The distance metric used for edges is related to the
// time-lock+fee costs along a particular edge. If a path is found, this
// function returns a slice of ChannelHop structs which encoded the chosen path
// (backwards) from the target to the source.
func findRoute(graph *channeldb.ChannelGraph, sourceNode *channeldb.LightningNode,
target *btcec.PublicKey, amt btcutil.Amount) ([]*ChannelHop, error) {
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// First we'll initialize an empty heap which'll help us to quickly
// locate the next edge we should visit next during our graph
// traversal.
var nodeHeap distanceHeap
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// For each node/vertex the graph we create an entry in the distance
// map for the node set with a distance of "infinity". We also mark
// add the node to our set of unvisited nodes.
distance := make(map[vertex]nodeWithDist)
if err := graph.ForEachNode(func(node *channeldb.LightningNode) error {
// TODO(roasbeef): with larger graph can just use disk seeks
// with a visited map
distance[newVertex(node.PubKey)] = nodeWithDist{
dist: infinity,
node: node,
}
return nil
}); err != nil {
return nil, err
}
// To start, we add the source of our path finding attempt to the
// distance map with with a distance of 0. This indicates our starting
// point in the graph traversal.
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
sourceVertex := newVertex(sourceNode.PubKey)
distance[sourceVertex] = nodeWithDist{
dist: 0,
node: sourceNode,
}
// To start, our source node will the sole item within our distance
// heap.
heap.Push(&nodeHeap, distance[sourceVertex])
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// We'll use this map as a series of "previous" hop pointers. So to get
// to `vertex` we'll take the edge that it's mapped to within `prev`.
prev := make(map[vertex]edgeWithPrev)
for nodeHeap.Len() != 0 {
// Fetch the node within the smallest distance from our source
// from the heap.
bestNode := heap.Pop(&nodeHeap).(nodeWithDist).node
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// If we've reached our target (or we don't have any outgoing
// edges), then we're done here and can exit the graph
// traversal early.
if bestNode.PubKey.IsEqual(target) {
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
break
}
// Now that we've found the next potential step to take we'll
// examine all the outgoing edge (channels) from this node to
// further our graph traversal.
pivot := newVertex(bestNode.PubKey)
err := bestNode.ForEachChannel(nil, func(edgeInfo *channeldb.ChannelEdgeInfo,
edge *channeldb.ChannelEdgePolicy) error {
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
// Compute the tentative distance to this new
// channel/edge which is the distance to our current
// pivot node plus the weight of this edge.
tempDist := distance[pivot].dist + edgeWeight(edge)
// If this new tentative distance is better than the
// current best known distance to this node, then we
// record the new better distance, and also populate
// our "next hop" map with this edge. We'll also shave
// off irrelevant edges by adding the sufficient
// capacity of an edge to our relaxation condition.
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
v := newVertex(edge.Node.PubKey)
if tempDist < distance[v].dist &&
edgeInfo.Capacity >= amt {
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
distance[v] = nodeWithDist{
dist: tempDist,
node: edge.Node,
}
prev[v] = edgeWithPrev{
edge: &ChannelHop{
ChannelEdgePolicy: edge,
Capacity: edgeInfo.Capacity,
},
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
prevNode: bestNode.PubKey,
}
// Add this new node to our heap as we'd like
// to further explore down this edge.
heap.Push(&nodeHeap, distance[v])
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
}
return nil
})
if err != nil {
return nil, err
}
}
// If the target node isn't found in the prev hop map, then a path
// doesn't exist, so we terminate in an error.
if _, ok := prev[newVertex(target)]; !ok {
return nil, ErrNoPathFound
}
// If the potential route if below the max hop limit, then we'll use
// the prevHop map to unravel the path. We end up with a list of edges
// in the reverse direction which we'll use to properly calculate the
// timelock and fee values.
pathEdges := make([]*ChannelHop, 0, len(prev))
prevNode := newVertex(target)
for prevNode != sourceVertex { // TODO(roasbeef): assumes no cycles
// Add the current hop to the limit of path edges then walk
// backwards from this hop via the prev pointer for this hop
// within the prevHop map.
pathEdges = append(pathEdges, prev[prevNode].edge)
prevNode = newVertex(prev[prevNode].prevNode)
}
// The route is invalid if it spans more than 20 hops. The current
// Sphinx (onion routing) implementation can only encode up to 20 hops
// as the entire packet is fixed size. If this route is more than 20
// hops, then it's invalid.
if len(pathEdges) > HopLimit {
return nil, ErrMaxHopsExceeded
}
return pathEdges, nil
}
routing: rewrite package to conform to BOLT07 and factor in fees+timelocks This commit overhauls the routing package significantly to simplify the code, conform to the rest of the coding style within the package, and observe the new authenticated gossiping scheme outlined in BOLT07. As a major step towards a more realistic path finding algorithm, fees are properly calculated and observed during path finding. If a path has sufficient capacity _before_ fees are applied, but afterwards the finalized route would exceed the capacity of a single link, the route is marked as invalid. Currently a naive weighting algorithm is used which only factors in the time-lock delta at each hop, thereby optimizing for the lowest time lock. Fee calculation also isn’t finalized since we aren’t yet using milli-satoshi throughout the daemon. The final TODO item within the PR is to properly perform a multi-path search and rank the results based on a summation heuristic rather than just return the first (out of many) route found. On the server side, once nodes are initially connected to the daemon, our routing table will be synced with the peer’s using a naive “just send everything scheme” to hold us over until I spec out some a efficient graph reconciliation protocol. Additionally, the routing table is now pruned by the channel router itself once new blocks arrive rather than depending on peers to tell us when a channel flaps or is closed. Finally, the validation of peer announcements aren’t yet fully implemented as they’ll be implemented within the pending discovery package that was blocking on the completion of this package. Most off the routing message processing will be moved out of this package and into the discovery package where full validation will be carried out.
2016-12-26 21:20:26 -08:00
}