lnd.xprv/routing/router.go

2320 lines
76 KiB
Go
Raw Normal View History

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.
2016-12-27 08:20:26 +03:00
package routing
import (
"bytes"
"crypto/sha256"
"fmt"
"runtime"
"sort"
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.
2016-12-27 08:20:26 +03:00
"sync"
"sync/atomic"
"time"
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.
2016-12-27 08:20:26 +03:00
"github.com/btcsuite/btcd/btcec"
"github.com/btcsuite/btcd/wire"
"github.com/btcsuite/btcutil"
"github.com/coreos/bbolt"
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.
2016-12-27 08:20:26 +03:00
"github.com/davecgh/go-spew/spew"
"github.com/go-errors/errors"
"github.com/lightningnetwork/lightning-onion"
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.
2016-12-27 08:20:26 +03:00
"github.com/lightningnetwork/lnd/channeldb"
"github.com/lightningnetwork/lnd/htlcswitch"
"github.com/lightningnetwork/lnd/input"
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.
2016-12-27 08:20:26 +03:00
"github.com/lightningnetwork/lnd/lnwallet"
"github.com/lightningnetwork/lnd/lnwire"
"github.com/lightningnetwork/lnd/multimutex"
"github.com/lightningnetwork/lnd/routing/chainview"
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.
2016-12-27 08:20:26 +03:00
)
const (
// DefaultFinalCLTVDelta is the default value to be used as the final
// CLTV delta for a route if one is unspecified.
DefaultFinalCLTVDelta = 9
// defaultPayAttemptTimeout is a duration that we'll use to determine
// if we should give up on a payment attempt. This will be used if a
// value isn't specified in the LightningNode struct.
defaultPayAttemptTimeout = time.Duration(time.Second * 60)
)
var (
// ErrNoRouteHopsProvided is returned when a caller attempts to
// construct a new sphinx packet, but provides an empty set of hops for
// each route.
ErrNoRouteHopsProvided = fmt.Errorf("empty route hops provided")
)
// ChannelGraphSource represents the source of information about the topology
// of the lightning network. It's responsible for the addition of nodes, edges,
// applying edge updates, and returning the current block height with which the
// topology is synchronized.
type ChannelGraphSource interface {
// AddNode is used to add information about a node to the router
// database. If the node with this pubkey is not present in an existing
// channel, it will be ignored.
AddNode(node *channeldb.LightningNode) error
// AddEdge is used to add edge/channel to the topology of the router,
// after all information about channel will be gathered this
// edge/channel might be used in construction of payment path.
AddEdge(edge *channeldb.ChannelEdgeInfo) error
// AddProof updates the channel edge info with proof which is needed to
// properly announce the edge to the rest of the network.
AddProof(chanID lnwire.ShortChannelID, proof *channeldb.ChannelAuthProof) error
// UpdateEdge is used to update edge information, without this message
// edge considered as not fully constructed.
UpdateEdge(policy *channeldb.ChannelEdgePolicy) error
// IsStaleNode returns true if the graph source has a node announcement
// for the target node with a more recent timestamp. This method will
// also return true if we don't have an active channel announcement for
// the target node.
IsStaleNode(node Vertex, timestamp time.Time) bool
// IsPublicNode determines whether the given vertex is seen as a public
// node in the graph from the graph's source node's point of view.
IsPublicNode(node Vertex) (bool, error)
// IsKnownEdge returns true if the graph source already knows of the
// passed channel ID.
IsKnownEdge(chanID lnwire.ShortChannelID) bool
// IsStaleEdgePolicy returns true if the graph source has a channel
// edge for the passed channel ID (and flags) that have a more recent
// timestamp.
IsStaleEdgePolicy(chanID lnwire.ShortChannelID, timestamp time.Time,
flags lnwire.ChanUpdateChanFlags) bool
// ForAllOutgoingChannels is used to iterate over all channels
// emanating from the "source" node which is the center of the
// star-graph.
ForAllOutgoingChannels(cb func(c *channeldb.ChannelEdgeInfo,
e *channeldb.ChannelEdgePolicy) error) error
// CurrentBlockHeight returns the block height from POV of the router
// subsystem.
CurrentBlockHeight() (uint32, error)
// GetChannelByID return the channel by the channel id.
GetChannelByID(chanID lnwire.ShortChannelID) (*channeldb.ChannelEdgeInfo,
*channeldb.ChannelEdgePolicy, *channeldb.ChannelEdgePolicy, error)
// FetchLightningNode attempts to look up a target node by its identity
// public key. channeldb.ErrGraphNodeNotFound is returned if the node
// doesn't exist within the graph.
FetchLightningNode(Vertex) (*channeldb.LightningNode, error)
// ForEachNode is used to iterate over every node in the known graph.
ForEachNode(func(node *channeldb.LightningNode) error) error
// ForEachChannel is used to iterate over every channel in the known
// graph.
ForEachChannel(func(chanInfo *channeldb.ChannelEdgeInfo,
e1, e2 *channeldb.ChannelEdgePolicy) error) error
}
// FeeSchema is the set fee configuration for a Lightning Node on the network.
// Using the coefficients described within the schema, the required fee to
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.
2016-12-27 08:20:26 +03:00
// forward outgoing payments can be derived.
type FeeSchema struct {
// BaseFee is the base amount of milli-satoshis that will be chained
// for ANY payment forwarded.
BaseFee lnwire.MilliSatoshi
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.
2016-12-27 08:20:26 +03:00
// FeeRate is the rate that will be charged for forwarding payments.
// This value should be interpreted as the numerator for a fraction
// (fixed point arithmetic) whose denominator is 1 million. As a result
// the effective fee rate charged per mSAT will be: (amount *
// FeeRate/1,000,000).
FeeRate uint32
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.
2016-12-27 08:20:26 +03:00
}
// ChannelPolicy holds the parameters that determine the policy we enforce
// when forwarding payments on a channel. These parameters are communicated
// to the rest of the network in ChannelUpdate messages.
type ChannelPolicy struct {
// FeeSchema holds the fee configuration for a channel.
FeeSchema
// TimeLockDelta is the required HTLC timelock delta to be used
// when forwarding payments.
TimeLockDelta uint32
}
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.
2016-12-27 08:20:26 +03:00
// Config defines the configuration for the ChannelRouter. ALL elements within
// the configuration MUST be non-nil for the ChannelRouter to carry out its
// duties.
type Config struct {
// Graph is the channel graph that the ChannelRouter will use to gather
// metrics from and also to carry out path finding queries.
// TODO(roasbeef): make into an interface
Graph *channeldb.ChannelGraph
// Chain is the router's source to the most up-to-date blockchain data.
// All incoming advertised channels will be checked against the chain
// to ensure that the channels advertised are still open.
Chain lnwallet.BlockChainIO
// ChainView is an instance of a FilteredChainView which is used to
// watch the sub-set of the UTXO set (the set of active channels) that
// we need in order to properly maintain the channel graph.
ChainView chainview.FilteredChainView
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.
2016-12-27 08:20:26 +03:00
// SendToSwitch is a function that directs a link-layer switch to
// forward a fully encoded payment to the first hop in the route
// denoted by its public key. A non-nil error is to be returned if the
// payment was unsuccessful.
SendToSwitch func(firstHop lnwire.ShortChannelID,
htlcAdd *lnwire.UpdateAddHTLC,
circuit *sphinx.Circuit) ([sha256.Size]byte, error)
// ChannelPruneExpiry is the duration used to determine if a channel
// should be pruned or not. If the delta between now and when the
// channel was last updated is greater than ChannelPruneExpiry, then
// the channel is marked as a zombie channel eligible for pruning.
ChannelPruneExpiry time.Duration
// GraphPruneInterval is used as an interval to determine how often we
// should examine the channel graph to garbage collect zombie channels.
GraphPruneInterval time.Duration
// QueryBandwidth is a method that allows the router to query the lower
// link layer to determine the up to date available bandwidth at a
// prospective link to be traversed. If the link isn't available, then
// a value of zero should be returned. Otherwise, the current up to
// date knowledge of the available bandwidth of the link should be
// returned.
QueryBandwidth func(edge *channeldb.ChannelEdgeInfo) lnwire.MilliSatoshi
2018-08-30 05:05:13 +03:00
// AssumeChannelValid toggles whether or not the router will check for
// spentness of channel outpoints. For neutrino, this saves long rescans
// from blocking initial usage of the wallet. This should only be
// enabled on testnet.
AssumeChannelValid bool
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.
2016-12-27 08:20:26 +03:00
}
// routeTuple is an entry within the ChannelRouter's route cache. We cache
// prospective routes based on first the destination, and then the target
// amount. We required the target amount as that will influence the available
// set of paths for a payment.
type routeTuple struct {
amt lnwire.MilliSatoshi
dest [33]byte
}
// newRouteTuple creates a new route tuple from the target and amount.
func newRouteTuple(amt lnwire.MilliSatoshi, dest []byte) routeTuple {
r := routeTuple{
amt: amt,
}
copy(r.dest[:], dest)
return r
}
// edgeLocator is a struct used to identify a specific edge. The direction
// fields takes the value of 0 or 1 and is identical in definition to the
// channel direction flag. A value of 0 means the direction from the lower node
// pubkey to the higher.
type edgeLocator struct {
channelID uint64
direction uint8
}
// newEdgeLocatorByPubkeys returns an edgeLocator based on its end point
// pubkeys.
func newEdgeLocatorByPubkeys(channelID uint64, fromNode, toNode *Vertex) *edgeLocator {
// Determine direction based on lexicographical ordering of both
// pubkeys.
var direction uint8
if bytes.Compare(fromNode[:], toNode[:]) == 1 {
direction = 1
}
return &edgeLocator{
channelID: channelID,
direction: direction,
}
}
// newEdgeLocator extracts an edgeLocator based for a full edge policy
// structure.
func newEdgeLocator(edge *channeldb.ChannelEdgePolicy) *edgeLocator {
return &edgeLocator{
channelID: edge.ChannelID,
direction: uint8(edge.ChannelFlags & lnwire.ChanUpdateDirection),
}
}
// String returns a human readable version of the edgeLocator values.
func (e *edgeLocator) String() string {
return fmt.Sprintf("%v:%v", e.channelID, e.direction)
}
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.
2016-12-27 08:20:26 +03:00
// ChannelRouter is the layer 3 router within the Lightning stack. Below the
// ChannelRouter is the HtlcSwitch, and below that is the Bitcoin blockchain
// itself. The primary role of the ChannelRouter is to respond to queries for
// potential routes that can support a payment amount, and also general graph
// reachability questions. The router will prune the channel graph
// automatically as new blocks are discovered which spend certain known funding
// outpoints, thereby closing their respective channels.
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.
2016-12-27 08:20:26 +03:00
type ChannelRouter struct {
ntfnClientCounter uint64 // To be used atomically.
started uint32 // To be used atomically.
stopped uint32 // To be used atomically.
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.
2016-12-27 08:20:26 +03:00
bestHeight uint32 // To be used atomically.
// cfg is a copy of the configuration struct that the ChannelRouter was
// initialized with.
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.
2016-12-27 08:20:26 +03:00
cfg *Config
// selfNode is the center of the star-graph centered around the
// ChannelRouter. The ChannelRouter uses this node as a starting point
// when doing any path finding.
selfNode *channeldb.LightningNode
// routeCache is a map that caches the k-shortest paths from ourselves
// to a given target destination for a particular payment amount. This
// map is used as an optimization to speed up subsequent payments to a
// particular destination. This map will be cleared each time a new
// channel announcement is accepted, or a new block arrives that
// results in channels being closed.
//
// TODO(roasbeef): make LRU
routeCacheMtx sync.RWMutex
routeCache map[routeTuple][]*Route
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.
2016-12-27 08:20:26 +03:00
// newBlocks is a channel in which new blocks connected to the end of
// the main chain are sent over, and blocks updated after a call to
// UpdateFilter.
newBlocks <-chan *chainview.FilteredBlock
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.
2016-12-27 08:20:26 +03:00
// staleBlocks is a channel in which blocks disconnected fromt the end
// of our currently known best chain are sent over.
staleBlocks <-chan *chainview.FilteredBlock
// networkUpdates is a channel that carries new topology updates
// messages from outside the ChannelRouter to be processed by the
// networkHandler.
networkUpdates chan *routingMsg
// topologyClients maps a client's unique notification ID to a
// topologyClient client that contains its notification dispatch
// channel.
topologyClients map[uint64]*topologyClient
// ntfnClientUpdates is a channel that's used to send new updates to
// topology notification clients to the ChannelRouter. Updates either
// add a new notification client, or cancel notifications for an
// existing client.
ntfnClientUpdates chan *topologyClientUpdate
// missionControl is a shared memory of sorts that executions of
// payment path finding use in order to remember which vertexes/edges
// were pruned from prior attempts. During SendPayment execution,
// errors sent by nodes are mapped into a vertex or edge to be pruned.
// Each run will then take into account this set of pruned
// vertexes/edges to reduce route failure and pass on graph information
// gained to the next execution.
missionControl *missionControl
// channelEdgeMtx is a mutex we use to make sure we process only one
// ChannelEdgePolicy at a time for a given channelID, to ensure
// consistency between the various database accesses.
channelEdgeMtx *multimutex.Mutex
rejectMtx sync.RWMutex
rejectCache map[uint64]struct{}
sync.RWMutex
quit chan struct{}
wg sync.WaitGroup
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.
2016-12-27 08:20:26 +03:00
}
// A compile time check to ensure ChannelRouter implements the
// ChannelGraphSource interface.
var _ ChannelGraphSource = (*ChannelRouter)(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.
2016-12-27 08:20:26 +03:00
// New creates a new instance of the ChannelRouter with the specified
// configuration parameters. As part of initialization, if the router detects
// that the channel graph isn't fully in sync with the latest UTXO (since the
// channel graph is a subset of the UTXO set) set, then the router will proceed
// to fully sync to the latest state of the UTXO set.
func New(cfg Config) (*ChannelRouter, error) {
selfNode, err := cfg.Graph.SourceNode()
if err != nil {
return nil, err
}
r := &ChannelRouter{
cfg: &cfg,
networkUpdates: make(chan *routingMsg),
topologyClients: make(map[uint64]*topologyClient),
ntfnClientUpdates: make(chan *topologyClientUpdate),
channelEdgeMtx: multimutex.NewMutex(),
selfNode: selfNode,
routeCache: make(map[routeTuple][]*Route),
rejectCache: make(map[uint64]struct{}),
quit: make(chan struct{}),
}
r.missionControl = newMissionControl(
cfg.Graph, selfNode, cfg.QueryBandwidth,
)
return r, 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.
2016-12-27 08:20:26 +03:00
}
// Start launches all the goroutines the ChannelRouter requires to carry out
// its duties. If the router has already been started, then this method is a
// noop.
func (r *ChannelRouter) Start() error {
if !atomic.CompareAndSwapUint32(&r.started, 0, 1) {
return nil
}
log.Tracef("Channel Router starting")
// First, we'll start the chain view instance (if it isn't already
// started).
if err := r.cfg.ChainView.Start(); err != 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.
2016-12-27 08:20:26 +03:00
return err
}
// Once the instance is active, we'll fetch the channel we'll receive
// notifications over.
r.newBlocks = r.cfg.ChainView.FilteredBlocks()
r.staleBlocks = r.cfg.ChainView.DisconnectedBlocks()
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.
2016-12-27 08:20:26 +03:00
bestHash, bestHeight, err := r.cfg.Chain.GetBestBlock()
if err != nil {
return err
}
if _, _, err := r.cfg.Graph.PruneTip(); err != nil {
switch {
// If the graph has never been pruned, or hasn't fully been
// created yet, then we don't treat this as an explicit error.
case err == channeldb.ErrGraphNeverPruned:
fallthrough
case err == channeldb.ErrGraphNotFound:
// If the graph has never been pruned, then we'll set
// the prune height to the current best height of the
// chain backend.
_, err = r.cfg.Graph.PruneGraph(
nil, bestHash, uint32(bestHeight),
)
if err != nil {
return err
}
default:
return err
}
}
// Before we perform our manual block pruning, we'll construct and
// apply a fresh chain filter to the active FilteredChainView instance.
// We do this before, as otherwise we may miss on-chain events as the
// filter hasn't properly been applied.
channelView, err := r.cfg.Graph.ChannelView()
if err != nil && err != channeldb.ErrGraphNoEdgesFound {
return err
}
log.Infof("Filtering chain using %v channels active", len(channelView))
if len(channelView) != 0 {
err = r.cfg.ChainView.UpdateFilter(
channelView, uint32(bestHeight),
)
if err != nil {
return err
}
}
// Before we begin normal operation of the router, we first need to
// synchronize the channel graph to the latest state of the UTXO set.
if err := r.syncGraphWithChain(); err != nil {
return err
}
// Finally, before we proceed, we'll prune any unconnected nodes from
// the graph in order to ensure we maintain a tight graph of "useful"
// nodes.
err = r.cfg.Graph.PruneGraphNodes()
if err != nil && err != channeldb.ErrGraphNodesNotFound {
return err
}
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.
2016-12-27 08:20:26 +03:00
r.wg.Add(1)
go r.networkHandler()
return nil
}
// Stop signals the ChannelRouter to gracefully halt all routines. This method
// will *block* until all goroutines have excited. If the channel router has
// already stopped then this method will return immediately.
func (r *ChannelRouter) Stop() error {
if !atomic.CompareAndSwapUint32(&r.stopped, 0, 1) {
return nil
}
log.Infof("Channel Router shutting down")
if err := r.cfg.ChainView.Stop(); err != nil {
return err
}
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.
2016-12-27 08:20:26 +03:00
close(r.quit)
r.wg.Wait()
return nil
}
// syncGraphWithChain attempts to synchronize the current channel graph with
// the latest UTXO set state. This process involves pruning from the channel
// graph any channels which have been closed by spending their funding output
// since we've been down.
func (r *ChannelRouter) syncGraphWithChain() error {
// First, we'll need to check to see if we're already in sync with the
// latest state of the UTXO set.
bestHash, bestHeight, err := r.cfg.Chain.GetBestBlock()
if err != nil {
return err
}
r.bestHeight = uint32(bestHeight)
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.
2016-12-27 08:20:26 +03:00
pruneHash, pruneHeight, err := r.cfg.Graph.PruneTip()
if err != nil {
switch {
// If the graph has never been pruned, or hasn't fully been
// created yet, then we don't treat this as an explicit error.
case err == channeldb.ErrGraphNeverPruned:
case err == channeldb.ErrGraphNotFound:
default:
return err
}
}
log.Infof("Prune tip for Channel Graph: height=%v, hash=%v", pruneHeight,
pruneHash)
switch {
// If the graph has never been pruned, then we can exit early as this
// entails it's being created for the first time and hasn't seen any
// block or created channels.
case pruneHeight == 0 || pruneHash == nil:
return nil
// If the block hashes and heights match exactly, then we don't need to
// prune the channel graph as we're already fully in sync.
case bestHash.IsEqual(pruneHash) && uint32(bestHeight) == pruneHeight:
return nil
}
// If the main chain blockhash at prune height is different from the
// prune hash, this might indicate the database is on a stale branch.
mainBlockHash, err := r.cfg.Chain.GetBlockHash(int64(pruneHeight))
if err != nil {
return err
}
// While we are on a stale branch of the chain, walk backwards to find
// first common block.
for !pruneHash.IsEqual(mainBlockHash) {
log.Infof("channel graph is stale. Disconnecting block %v "+
"(hash=%v)", pruneHeight, pruneHash)
// Prune the graph for every channel that was opened at height
// >= pruneHeight.
_, err := r.cfg.Graph.DisconnectBlockAtHeight(pruneHeight)
if err != nil {
return err
}
pruneHash, pruneHeight, err = r.cfg.Graph.PruneTip()
if err != nil {
switch {
// If at this point the graph has never been pruned, we
// can exit as this entails we are back to the point
// where it hasn't seen any block or created channels,
// alas there's nothing left to prune.
case err == channeldb.ErrGraphNeverPruned:
return nil
case err == channeldb.ErrGraphNotFound:
return nil
default:
return err
}
}
mainBlockHash, err = r.cfg.Chain.GetBlockHash(int64(pruneHeight))
if err != nil {
return err
}
}
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.
2016-12-27 08:20:26 +03:00
log.Infof("Syncing channel graph from height=%v (hash=%v) to height=%v "+
"(hash=%v)", pruneHeight, pruneHash, bestHeight, bestHash)
// If we're not yet caught up, then we'll walk forward in the chain in
// the chain pruning the channel graph with each new block in the chain
// that hasn't yet been consumed by the channel graph.
var numChansClosed uint32
for nextHeight := pruneHeight + 1; nextHeight <= uint32(bestHeight); nextHeight++ {
// Using the next height, request a manual block pruning from
// the chainview for the particular block hash.
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.
2016-12-27 08:20:26 +03:00
nextHash, err := r.cfg.Chain.GetBlockHash(int64(nextHeight))
if err != nil {
return err
}
filterBlock, err := r.cfg.ChainView.FilterBlock(nextHash)
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.
2016-12-27 08:20:26 +03:00
if err != nil {
return err
}
// We're only interested in all prior outputs that have been
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.
2016-12-27 08:20:26 +03:00
// spent in the block, so collate all the referenced previous
// outpoints within each tx and input.
var spentOutputs []*wire.OutPoint
for _, tx := range filterBlock.Transactions {
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.
2016-12-27 08:20:26 +03:00
for _, txIn := range tx.TxIn {
spentOutputs = append(spentOutputs,
&txIn.PreviousOutPoint)
}
}
// With the spent outputs gathered, attempt to prune the
// channel graph, also passing in the hash+height of the block
// being pruned so the prune tip can be updated.
closedChans, err := r.cfg.Graph.PruneGraph(spentOutputs,
nextHash,
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.
2016-12-27 08:20:26 +03:00
nextHeight)
if err != nil {
return err
}
numClosed := uint32(len(closedChans))
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.
2016-12-27 08:20:26 +03:00
log.Infof("Block %v (height=%v) closed %v channels",
nextHash, nextHeight, numClosed)
numChansClosed += numClosed
}
log.Infof("Graph pruning complete: %v channels were closed since "+
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.
2016-12-27 08:20:26 +03:00
"height %v", numChansClosed, pruneHeight)
return nil
}
// pruneZombieChans is a method that will be called periodically to prune out
// any "zombie" channels. We consider channels zombies if *both* edges haven't
// been updated since our zombie horizon. We do this periodically to keep a
// health, lively routing table.
func (r *ChannelRouter) pruneZombieChans() error {
var chansToPrune []wire.OutPoint
chanExpiry := r.cfg.ChannelPruneExpiry
log.Infof("Examining Channel Graph for zombie channels")
// First, we'll collect all the channels which are eligible for garbage
// collection due to being zombies.
filterPruneChans := func(info *channeldb.ChannelEdgeInfo,
e1, e2 *channeldb.ChannelEdgePolicy) error {
// We'll ensure that we don't attempt to prune our *own*
// channels from the graph, as in any case this should be
// re-advertised by the sub-system above us.
if info.NodeKey1Bytes == r.selfNode.PubKeyBytes ||
info.NodeKey2Bytes == r.selfNode.PubKeyBytes {
return nil
}
// If *both* edges haven't been updated for a period of
// chanExpiry, then we'll mark the channel itself as eligible
// for graph pruning.
e1Zombie, e2Zombie := true, true
if e1 != nil {
e1Zombie = time.Since(e1.LastUpdate) >= chanExpiry
if e1Zombie {
log.Tracef("Edge #1 of ChannelPoint(%v) "+
"last update: %v",
info.ChannelPoint, e1.LastUpdate)
}
}
if e2 != nil {
e2Zombie = time.Since(e2.LastUpdate) >= chanExpiry
if e2Zombie {
log.Tracef("Edge #2 of ChannelPoint(%v) "+
"last update: %v",
info.ChannelPoint, e2.LastUpdate)
}
}
if e1Zombie && e2Zombie {
log.Debugf("ChannelPoint(%v) is a zombie, collecting "+
"to prune", info.ChannelPoint)
// TODO(roasbeef): add ability to delete single
// directional edge
chansToPrune = append(chansToPrune, info.ChannelPoint)
// As we're detecting this as a zombie channel, we'll
// add this to the set of recently rejected items so we
// don't re-accept it shortly after.
r.rejectCache[info.ChannelID] = struct{}{}
}
return nil
}
r.rejectMtx.Lock()
defer r.rejectMtx.Unlock()
err := r.cfg.Graph.ForEachChannel(filterPruneChans)
if err != nil {
return fmt.Errorf("Unable to filter local zombie "+
"chans: %v", err)
}
log.Infof("Pruning %v Zombie Channels", len(chansToPrune))
// With the set zombie-like channels obtained, we'll do another pass to
// delete al zombie channels from the channel graph.
for _, chanToPrune := range chansToPrune {
log.Tracef("Pruning zombie chan ChannelPoint(%v)", chanToPrune)
err := r.cfg.Graph.DeleteChannelEdge(&chanToPrune)
if err != nil {
return fmt.Errorf("Unable to prune zombie "+
"chans: %v", err)
}
}
return 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.
2016-12-27 08:20:26 +03:00
// networkHandler is the primary goroutine for the ChannelRouter. The roles of
// this goroutine include answering queries related to the state of the
// network, pruning the graph on new block notification, applying network
// updates, and registering new topology clients.
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.
2016-12-27 08:20:26 +03:00
//
// NOTE: This MUST be run as a goroutine.
func (r *ChannelRouter) networkHandler() {
defer r.wg.Done()
graphPruneTicker := time.NewTicker(r.cfg.GraphPruneInterval)
defer graphPruneTicker.Stop()
// We'll use this validation barrier to ensure that we process all jobs
// in the proper order during parallel validation.
validationBarrier := NewValidationBarrier(runtime.NumCPU()*4, r.quit)
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.
2016-12-27 08:20:26 +03:00
for {
select {
// A new fully validated network update has just arrived. As a
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.
2016-12-27 08:20:26 +03:00
// result we'll modify the channel graph accordingly depending
// on the exact type of the message.
case update := <-r.networkUpdates:
// We'll set up any dependants, and wait until a free
// slot for this job opens up, this allow us to not
// have thousands of goroutines active.
validationBarrier.InitJobDependencies(update.msg)
r.wg.Add(1)
go func() {
defer r.wg.Done()
defer validationBarrier.CompleteJob()
// If this message has an existing dependency,
// then we'll wait until that has been fully
// validated before we proceed.
err := validationBarrier.WaitForDependants(
update.msg,
)
if err != nil {
if err != ErrVBarrierShuttingDown {
log.Warnf("unexpected error "+
"during validation "+
"barrier shutdown: %v",
err)
}
return
}
// Process the routing update to determine if
// this is either a new update from our PoV or
// an update to a prior vertex/edge we
// previously accepted.
err = r.processUpdate(update.msg)
update.err <- err
// If this message had any dependencies, then
// we can now signal them to continue.
validationBarrier.SignalDependants(update.msg)
if err != nil {
return
}
// Send off a new notification for the newly
// accepted update.
topChange := &TopologyChange{}
err = addToTopologyChange(
r.cfg.Graph, topChange, update.msg,
)
if err != nil {
log.Errorf("unable to update topology "+
"change notification: %v", err)
return
}
if !topChange.isEmpty() {
r.notifyTopologyChange(topChange)
}
}()
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.
2016-12-27 08:20:26 +03:00
// TODO(roasbeef): remove all unconnected vertexes
// after N blocks pass with no corresponding
// announcements.
case chainUpdate, ok := <-r.staleBlocks:
// If the channel has been closed, then this indicates
// the daemon is shutting down, so we exit ourselves.
if !ok {
return
}
// Since this block is stale, we update our best height
// to the previous block.
blockHeight := uint32(chainUpdate.Height)
atomic.StoreUint32(&r.bestHeight, blockHeight-1)
// Update the channel graph to reflect that this block
// was disconnected.
_, err := r.cfg.Graph.DisconnectBlockAtHeight(blockHeight)
if err != nil {
log.Errorf("unable to prune graph with stale "+
"block: %v", err)
continue
}
// Invalidate the route cache, as some channels might
// not be confirmed anymore.
r.routeCacheMtx.Lock()
r.routeCache = make(map[routeTuple][]*Route)
r.routeCacheMtx.Unlock()
// TODO(halseth): notify client about the reorg?
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.
2016-12-27 08:20:26 +03:00
// A new block has arrived, so we can prune the channel graph
// of any channels which were closed in the block.
case chainUpdate, ok := <-r.newBlocks:
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.
2016-12-27 08:20:26 +03:00
// If the channel has been closed, then this indicates
// the daemon is shutting down, so we exit ourselves.
if !ok {
return
}
// We'll ensure that any new blocks received attach
// directly to the end of our main chain. If not, then
// we've somehow missed some blocks. We don't process
// this block as otherwise, we may miss on-chain
// events.
currentHeight := atomic.LoadUint32(&r.bestHeight)
if chainUpdate.Height != currentHeight+1 {
log.Errorf("out of order block: expecting "+
"height=%v, got height=%v", currentHeight+1,
chainUpdate.Height)
continue
}
// Once a new block arrives, we update our running
// track of the height of the chain tip.
blockHeight := uint32(chainUpdate.Height)
atomic.StoreUint32(&r.bestHeight, blockHeight)
log.Infof("Pruning channel graph using block %v (height=%v)",
chainUpdate.Hash, blockHeight)
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.
2016-12-27 08:20:26 +03:00
// We're only interested in all prior outputs that have
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.
2016-12-27 08:20:26 +03:00
// been spent in the block, so collate all the
// referenced previous outpoints within each tx and
// input.
var spentOutputs []*wire.OutPoint
for _, tx := range chainUpdate.Transactions {
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.
2016-12-27 08:20:26 +03:00
for _, txIn := range tx.TxIn {
spentOutputs = append(spentOutputs,
&txIn.PreviousOutPoint)
}
}
// With the spent outputs gathered, attempt to prune
// the channel graph, also passing in the hash+height
// of the block being pruned so the prune tip can be
// updated.
chansClosed, err := r.cfg.Graph.PruneGraph(spentOutputs,
&chainUpdate.Hash, chainUpdate.Height)
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.
2016-12-27 08:20:26 +03:00
if err != nil {
log.Errorf("unable to prune routing table: %v", err)
continue
}
log.Infof("Block %v (height=%v) closed %v channels",
chainUpdate.Hash, blockHeight, len(chansClosed))
// Invalidate the route cache as the block height has
// changed which will invalidate the HTLC timeouts we
// have crafted within each of the pre-computed routes.
//
// TODO(roasbeef): need to invalidate after each
// chan ann update?
// * can have map of chanID to routes involved, avoids
// full invalidation
r.routeCacheMtx.Lock()
r.routeCache = make(map[routeTuple][]*Route)
r.routeCacheMtx.Unlock()
if len(chansClosed) == 0 {
continue
}
// Notify all currently registered clients of the newly
// closed channels.
closeSummaries := createCloseSummaries(blockHeight, chansClosed...)
r.notifyTopologyChange(&TopologyChange{
ClosedChannels: closeSummaries,
})
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.
2016-12-27 08:20:26 +03:00
// A new notification client update has arrived. We're either
// gaining a new client, or cancelling notifications for an
// existing client.
case ntfnUpdate := <-r.ntfnClientUpdates:
clientID := ntfnUpdate.clientID
if ntfnUpdate.cancel {
r.RLock()
client, ok := r.topologyClients[ntfnUpdate.clientID]
r.RUnlock()
if ok {
r.Lock()
delete(r.topologyClients, clientID)
r.Unlock()
close(client.exit)
client.wg.Wait()
close(client.ntfnChan)
}
continue
}
r.Lock()
r.topologyClients[ntfnUpdate.clientID] = &topologyClient{
ntfnChan: ntfnUpdate.ntfnChan,
exit: make(chan struct{}),
}
r.Unlock()
// The graph prune ticker has ticked, so we'll examine the
// state of the known graph to filter out any zombie channels
// for pruning.
case <-graphPruneTicker.C:
if err := r.pruneZombieChans(); err != nil {
log.Errorf("unable to prune zombies: %v", err)
}
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.
2016-12-27 08:20:26 +03:00
// The router has been signalled to exit, to we exit our main
// loop so the wait group can be decremented.
case <-r.quit:
return
}
}
}
// assertNodeAnnFreshness returns a non-nil error if we have an announcement in
// the database for the passed node with a timestamp newer than the passed
// timestamp. ErrIgnored will be returned if we already have the node, and
// ErrOutdated will be returned if we have a timestamp that's after the new
// timestamp.
func (r *ChannelRouter) assertNodeAnnFreshness(node Vertex,
msgTimestamp time.Time) error {
// If we are not already aware of this node, it means that we don't
// know about any channel using this node. To avoid a DoS attack by
// node announcements, we will ignore such nodes. If we do know about
// this node, check that this update brings info newer than what we
// already have.
lastUpdate, exists, err := r.cfg.Graph.HasLightningNode(node)
if err != nil {
return errors.Errorf("unable to query for the "+
"existence of node: %v", err)
}
if !exists {
return newErrf(ErrIgnored, "Ignoring node announcement"+
" for node not found in channel graph (%x)",
node[:])
}
// If we've reached this point then we're aware of the vertex being
// advertised. So we now check if the new message has a new time stamp,
// if not then we won't accept the new data as it would override newer
// data.
if !lastUpdate.Before(msgTimestamp) {
return newErrf(ErrOutdated, "Ignoring outdated "+
"announcement for %x", node[:])
}
return nil
}
// processUpdate processes a new relate authenticated channel/edge, node or
// channel/edge update network update. If the update didn't affect the internal
// state of the draft due to either being out of date, invalid, or redundant,
// then error is returned.
func (r *ChannelRouter) processUpdate(msg interface{}) error {
var invalidateCache bool
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.
2016-12-27 08:20:26 +03:00
switch msg := msg.(type) {
case *channeldb.LightningNode:
// Before we add the node to the database, we'll check to see
// if the announcement is "fresh" or not. If it isn't, then
// we'll return an error.
err := r.assertNodeAnnFreshness(msg.PubKeyBytes, msg.LastUpdate)
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.
2016-12-27 08:20:26 +03:00
if err != nil {
return err
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.
2016-12-27 08:20:26 +03:00
}
if err := r.cfg.Graph.AddLightningNode(msg); err != nil {
return errors.Errorf("unable to add node %v to the "+
"graph: %v", msg.PubKeyBytes, err)
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.
2016-12-27 08:20:26 +03:00
}
log.Infof("Updated vertex data for node=%x", msg.PubKeyBytes)
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.
2016-12-27 08:20:26 +03:00
case *channeldb.ChannelEdgeInfo:
// If we recently rejected this channel edge, then we won't
// attempt to re-process it.
r.rejectMtx.RLock()
if _, ok := r.rejectCache[msg.ChannelID]; ok {
r.rejectMtx.RUnlock()
return newErrf(ErrRejected, "recently rejected "+
"chan_id=%v", msg.ChannelID)
}
r.rejectMtx.RUnlock()
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.
2016-12-27 08:20:26 +03:00
// Prior to processing the announcement we first check if we
// already know of this channel, if so, then we can exit early.
_, _, exists, err := r.cfg.Graph.HasChannelEdge(msg.ChannelID)
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.
2016-12-27 08:20:26 +03:00
if err != nil && err != channeldb.ErrGraphNoEdgesFound {
return errors.Errorf("unable to check for edge "+
"existence: %v", err)
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.
2016-12-27 08:20:26 +03:00
} else if exists {
return newErrf(ErrIgnored, "Ignoring msg for known "+
"chan_id=%v", msg.ChannelID)
}
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.
2016-12-27 08:20:26 +03:00
// Before we can add the channel to the channel graph, we need
// to obtain the full funding outpoint that's encoded within
// the channel ID.
channelID := lnwire.NewShortChanIDFromInt(msg.ChannelID)
2018-08-30 05:05:13 +03:00
fundingPoint, fundingTxOut, err := r.fetchChanPoint(&channelID)
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.
2016-12-27 08:20:26 +03:00
if err != nil {
r.rejectMtx.Lock()
r.rejectCache[msg.ChannelID] = struct{}{}
r.rejectMtx.Unlock()
return errors.Errorf("unable to fetch chan point for "+
"chan_id=%v: %v", msg.ChannelID, err)
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.
2016-12-27 08:20:26 +03:00
}
// Recreate witness output to be sure that declared in channel
// edge bitcoin keys and channel value corresponds to the
// reality.
witnessScript, err := input.GenMultiSigScript(
msg.BitcoinKey1Bytes[:], msg.BitcoinKey2Bytes[:],
)
if err != nil {
return err
}
fundingPkScript, err := input.WitnessScriptHash(witnessScript)
if err != nil {
return err
}
2018-08-30 05:05:13 +03:00
var chanUtxo *wire.TxOut
if r.cfg.AssumeChannelValid {
// If AssumeChannelValid is present, we'll just use the
// txout returned from fetchChanPoint.
chanUtxo = fundingTxOut
} else {
// Now that we have the funding outpoint of the channel,
// ensure that it hasn't yet been spent. If so, then
// this channel has been closed so we'll ignore it.
chanUtxo, err = r.cfg.Chain.GetUtxo(
fundingPoint, fundingPkScript,
channelID.BlockHeight,
)
if err != nil {
r.rejectMtx.Lock()
r.rejectCache[msg.ChannelID] = struct{}{}
r.rejectMtx.Unlock()
2018-08-30 05:05:13 +03:00
return errors.Errorf("unable to fetch utxo "+
"for chan_id=%v, chan_point=%v: %v",
msg.ChannelID, fundingPoint, err)
}
}
// By checking the equality of witness pkscripts we checks that
// funding witness script is multisignature lock which contains
// both local and remote public keys which was declared in
// channel edge and also that the announced channel value is
// right.
if !bytes.Equal(fundingPkScript, chanUtxo.PkScript) {
return errors.Errorf("pkScript mismatch: expected %x, "+
"got %x", fundingPkScript, chanUtxo.PkScript)
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.
2016-12-27 08:20:26 +03:00
}
// TODO(roasbeef): this is a hack, needs to be removed
// after commitment fees are dynamic.
msg.Capacity = btcutil.Amount(chanUtxo.Value)
msg.ChannelPoint = *fundingPoint
if err := r.cfg.Graph.AddChannelEdge(msg); err != nil {
return errors.Errorf("unable to add edge: %v", err)
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.
2016-12-27 08:20:26 +03:00
}
invalidateCache = true
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.
2016-12-27 08:20:26 +03:00
log.Infof("New channel discovered! Link "+
"connects %x and %x with ChannelPoint(%v): "+
"chan_id=%v, capacity=%v",
msg.NodeKey1Bytes, msg.NodeKey2Bytes,
fundingPoint, msg.ChannelID, msg.Capacity)
// As a new edge has been added to the channel graph, we'll
// update the current UTXO filter within our active
// FilteredChainView so we are notified if/when this channel is
// closed.
filterUpdate := []channeldb.EdgePoint{
{
FundingPkScript: fundingPkScript,
OutPoint: *fundingPoint,
},
}
err = r.cfg.ChainView.UpdateFilter(
filterUpdate, atomic.LoadUint32(&r.bestHeight),
)
if err != nil {
return errors.Errorf("unable to update chain "+
"view: %v", err)
}
case *channeldb.ChannelEdgePolicy:
// If we recently rejected this channel edge, then we won't
// attempt to re-process it.
r.rejectMtx.RLock()
if _, ok := r.rejectCache[msg.ChannelID]; ok {
r.rejectMtx.RUnlock()
return newErrf(ErrRejected, "recently rejected "+
"chan_id=%v", msg.ChannelID)
}
r.rejectMtx.RUnlock()
channelID := lnwire.NewShortChanIDFromInt(msg.ChannelID)
// We make sure to hold the mutex for this channel ID,
// such that no other goroutine is concurrently doing
// database accesses for the same channel ID.
r.channelEdgeMtx.Lock(msg.ChannelID)
defer r.channelEdgeMtx.Unlock(msg.ChannelID)
edge1Timestamp, edge2Timestamp, exists, err := r.cfg.Graph.HasChannelEdge(
msg.ChannelID,
)
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.
2016-12-27 08:20:26 +03:00
if err != nil && err != channeldb.ErrGraphNoEdgesFound {
return errors.Errorf("unable to check for edge "+
"existence: %v", err)
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.
2016-12-27 08:20:26 +03:00
}
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.
2016-12-27 08:20:26 +03:00
// As edges are directional edge node has a unique policy for
// the direction of the edge they control. Therefore we first
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.
2016-12-27 08:20:26 +03:00
// check if we already have the most up to date information for
// that edge. If this message has a timestamp not strictly
// newer than what we already know of we can exit early.
switch {
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.
2016-12-27 08:20:26 +03:00
// A flag set of 0 indicates this is an announcement for the
// "first" node in the channel.
case msg.ChannelFlags&lnwire.ChanUpdateDirection == 0:
// Ignore outdated message.
if !edge1Timestamp.Before(msg.LastUpdate) {
return newErrf(ErrOutdated, "Ignoring "+
"outdated update (flags=%v|%v) for "+
"known chan_id=%v", msg.MessageFlags,
msg.ChannelFlags, msg.ChannelID)
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.
2016-12-27 08:20:26 +03:00
}
// Similarly, a flag set of 1 indicates this is an announcement
// for the "second" node in the channel.
case msg.ChannelFlags&lnwire.ChanUpdateDirection == 1:
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.
2016-12-27 08:20:26 +03:00
// Ignore outdated message.
if !edge2Timestamp.Before(msg.LastUpdate) {
return newErrf(ErrOutdated, "Ignoring "+
"outdated update (flags=%v|%v) for "+
"known chan_id=%v", msg.MessageFlags,
msg.ChannelFlags, msg.ChannelID)
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.
2016-12-27 08:20:26 +03:00
}
}
2018-08-30 05:05:13 +03:00
if !exists && !r.cfg.AssumeChannelValid {
// Before we can update the channel information, we'll
// ensure that the target channel is still open by
// querying the utxo-set for its existence.
2018-08-30 05:05:13 +03:00
chanPoint, fundingTxOut, err := r.fetchChanPoint(
&channelID,
)
if err != nil {
r.rejectMtx.Lock()
r.rejectCache[msg.ChannelID] = struct{}{}
r.rejectMtx.Unlock()
return errors.Errorf("unable to fetch chan "+
"point for chan_id=%v: %v",
msg.ChannelID, err)
}
_, err = r.cfg.Chain.GetUtxo(
2018-08-30 05:05:13 +03:00
chanPoint, fundingTxOut.PkScript,
channelID.BlockHeight,
)
if err != nil {
r.rejectMtx.Lock()
r.rejectCache[msg.ChannelID] = struct{}{}
r.rejectMtx.Unlock()
return errors.Errorf("unable to fetch utxo for "+
"chan_id=%v: %v", msg.ChannelID, err)
}
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.
2016-12-27 08:20:26 +03:00
}
// Now that we know this isn't a stale update, we'll apply the
// new edge policy to the proper directional edge within the
// channel graph.
if err = r.cfg.Graph.UpdateEdgePolicy(msg); err != nil {
err := errors.Errorf("unable to add channel: %v", err)
log.Error(err)
return err
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.
2016-12-27 08:20:26 +03:00
}
invalidateCache = true
log.Tracef("New channel update applied: %v",
newLogClosure(func() string { return spew.Sdump(msg) }))
default:
return errors.Errorf("wrong routing update message type")
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.
2016-12-27 08:20:26 +03:00
}
// If we've received a channel update, then invalidate the route cache
// as channels within the graph have closed, which may affect our
// choice of the KSP's for a particular routeTuple.
if invalidateCache {
r.routeCacheMtx.Lock()
r.routeCache = make(map[routeTuple][]*Route)
r.routeCacheMtx.Unlock()
}
return 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.
2016-12-27 08:20:26 +03:00
}
// fetchChanPoint retrieves the original outpoint which is encoded within the
// channelID. This method also return the public key script for the target
// transaction.
//
// TODO(roasbeef): replace with call to GetBlockTransaction? (would allow to
// later use getblocktxn)
2018-08-30 05:05:13 +03:00
func (r *ChannelRouter) fetchChanPoint(
chanID *lnwire.ShortChannelID) (*wire.OutPoint, *wire.TxOut, 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.
2016-12-27 08:20:26 +03:00
// First fetch the block hash by the block number encoded, then use
// that hash to fetch the block itself.
blockNum := int64(chanID.BlockHeight)
blockHash, err := r.cfg.Chain.GetBlockHash(blockNum)
if err != nil {
return nil, nil, err
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.
2016-12-27 08:20:26 +03:00
}
fundingBlock, err := r.cfg.Chain.GetBlock(blockHash)
if err != nil {
return nil, nil, err
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.
2016-12-27 08:20:26 +03:00
}
// As a sanity check, ensure that the advertised transaction index is
// within the bounds of the total number of transactions within a
// block.
numTxns := uint32(len(fundingBlock.Transactions))
if chanID.TxIndex > numTxns-1 {
return nil, nil, fmt.Errorf("tx_index=#%v is out of range "+
"(max_index=%v), network_chan_id=%v\n", chanID.TxIndex,
numTxns-1, spew.Sdump(chanID))
}
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.
2016-12-27 08:20:26 +03:00
// Finally once we have the block itself, we seek to the targeted
2018-08-30 05:05:13 +03:00
// transaction index to obtain the funding output and txout.
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.
2016-12-27 08:20:26 +03:00
fundingTx := fundingBlock.Transactions[chanID.TxIndex]
2018-08-30 05:05:13 +03:00
outPoint := &wire.OutPoint{
Hash: fundingTx.TxHash(),
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.
2016-12-27 08:20:26 +03:00
Index: uint32(chanID.TxPosition),
2018-08-30 05:05:13 +03:00
}
txOut := fundingTx.TxOut[chanID.TxPosition]
return outPoint, txOut, 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.
2016-12-27 08:20:26 +03:00
}
// routingMsg couples a routing related routing topology update to the
// error channel.
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.
2016-12-27 08:20:26 +03:00
type routingMsg struct {
msg interface{}
err chan 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.
2016-12-27 08:20:26 +03:00
}
// pruneNodeFromRoutes accepts set of routes, and returns a new set of routes
// with the target node filtered out.
func pruneNodeFromRoutes(routes []*Route, skipNode Vertex) []*Route {
// TODO(roasbeef): pass in slice index?
prunedRoutes := make([]*Route, 0, len(routes))
for _, route := range routes {
if route.containsNode(skipNode) {
continue
}
prunedRoutes = append(prunedRoutes, route)
}
log.Tracef("Filtered out %v routes with node %x",
len(routes)-len(prunedRoutes), skipNode[:])
return prunedRoutes
}
// pruneChannelFromRoutes accepts a set of routes, and returns a new set of
// routes with the target channel filtered out.
func pruneChannelFromRoutes(routes []*Route, skipChan uint64) []*Route {
prunedRoutes := make([]*Route, 0, len(routes))
for _, route := range routes {
if route.containsChannel(skipChan) {
continue
}
prunedRoutes = append(prunedRoutes, route)
}
log.Tracef("Filtered out %v routes with channel %v",
len(routes)-len(prunedRoutes), skipChan)
return prunedRoutes
}
// pathsToFeeSortedRoutes takes a set of paths, and returns a corresponding set
2018-04-18 05:03:27 +03:00
// of routes. A route differs from a path in that it has full time-lock and
// fee information attached. The set of routes returned may be less than the
// initial set of paths as it's possible we drop a route if it can't handle the
// total payment flow after fees are calculated.
func pathsToFeeSortedRoutes(source Vertex, paths [][]*channeldb.ChannelEdgePolicy,
finalCLTVDelta uint16, amt, feeLimit lnwire.MilliSatoshi,
currentHeight uint32) ([]*Route, error) {
validRoutes := make([]*Route, 0, len(paths))
for _, path := range paths {
// Attempt to make the path into a route. We snip off the first
// hop in the path as it contains a "self-hop" that is inserted
// by our KSP algorithm.
route, err := newRoute(
amt, feeLimit, source, path[1:], currentHeight,
finalCLTVDelta,
)
if err != nil {
// TODO(roasbeef): report straw breaking edge?
continue
}
// If the path as enough total flow to support the computed
// route, then we'll add it to our set of valid routes.
validRoutes = append(validRoutes, route)
}
// If all our perspective routes were eliminating during the transition
// from path to route, then we'll return an error to the caller
if len(validRoutes) == 0 {
return nil, newErr(ErrNoPathFound, "unable to find a path to "+
"destination")
}
// Finally, we'll sort the set of validate routes to optimize for
// lowest total fees, using the required time-lock within the route as
// a tie-breaker.
sort.Slice(validRoutes, func(i, j int) bool {
// To make this decision we first check if the total fees
// required for both routes are equal. If so, then we'll let
// the total time lock be the tie breaker. Otherwise, we'll put
// the route with the lowest total fees first.
if validRoutes[i].TotalFees == validRoutes[j].TotalFees {
timeLockI := validRoutes[i].TotalTimeLock
timeLockJ := validRoutes[j].TotalTimeLock
return timeLockI < timeLockJ
}
return validRoutes[i].TotalFees < validRoutes[j].TotalFees
})
return validRoutes, nil
}
// FindRoutes attempts to query the ChannelRouter for a bounded number
// available paths to a particular target destination which is able to send
// `amt` after factoring in channel capacities and cumulative fees along each
2018-04-18 05:03:27 +03:00
// route. To `numPaths eligible paths, we use a modified version of
// Yen's algorithm which itself uses a modified version of Dijkstra's algorithm
// within its inner loop. Once we have a set of candidate routes, we calculate
// the required fee and time lock values running backwards along the route. The
// route that will be ranked the highest is the one with the lowest cumulative
// fee along the route.
func (r *ChannelRouter) FindRoutes(target *btcec.PublicKey,
amt, feeLimit lnwire.MilliSatoshi, numPaths uint32,
finalExpiry ...uint16) ([]*Route, error) {
var finalCLTVDelta uint16
if len(finalExpiry) == 0 {
finalCLTVDelta = DefaultFinalCLTVDelta
} else {
finalCLTVDelta = finalExpiry[0]
}
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.
2016-12-27 08:20:26 +03:00
dest := target.SerializeCompressed()
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.
2016-12-27 08:20:26 +03:00
log.Debugf("Searching for path to %x, sending %v", dest, amt)
// Before attempting to perform a series of graph traversals to find
// the k-shortest paths to the destination, we'll first consult our
// path cache
rt := newRouteTuple(amt, dest)
r.routeCacheMtx.RLock()
routes, ok := r.routeCache[rt]
r.routeCacheMtx.RUnlock()
// If we already have a cached route, and it contains at least the
// number of paths requested, then we'll return it directly as there's
// no need to repeat the computation.
if ok && uint32(len(routes)) >= numPaths {
return routes, nil
}
// If we don't have a set of routes cached, we'll query the graph for a
// set of potential routes to the destination node that can support our
// payment amount. If no such routes can be found then an error will be
// returned.
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.
2016-12-27 08:20:26 +03:00
// We can short circuit the routing by opportunistically checking to
// see if the target vertex event exists in the current graph.
targetVertex := NewVertex(target)
if _, exists, err := r.cfg.Graph.HasLightningNode(targetVertex); err != 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.
2016-12-27 08:20:26 +03:00
return nil, err
} else if !exists {
log.Debugf("Target %x is not in known graph", dest)
return nil, newErrf(ErrTargetNotInNetwork, "target not found")
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.
2016-12-27 08:20:26 +03:00
}
// We'll also fetch the current block height so we can properly
// calculate the required HTLC time locks within the route.
_, currentHeight, err := r.cfg.Chain.GetBestBlock()
if err != nil {
return nil, err
}
// Before we open the db transaction below, we'll attempt to obtain a
// set of bandwidth hints that can help us eliminate certain routes
// early on in the path finding process.
bandwidthHints, err := generateBandwidthHints(
r.selfNode, r.cfg.QueryBandwidth,
)
if err != nil {
return nil, err
}
tx, err := r.cfg.Graph.Database().Begin(false)
if err != nil {
tx.Rollback()
return nil, err
}
// Now that we know the destination is reachable within the graph,
// we'll execute our KSP algorithm to find the k-shortest paths from
// our source to the destination.
shortestPaths, err := findPaths(
2018-06-04 23:10:05 +03:00
tx, r.cfg.Graph, r.selfNode, target, amt, feeLimit, numPaths,
bandwidthHints,
)
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.
2016-12-27 08:20:26 +03:00
if err != nil {
tx.Rollback()
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.
2016-12-27 08:20:26 +03:00
return nil, err
}
tx.Rollback()
// Now that we have a set of paths, we'll need to turn them into
// *routes* by computing the required time-lock and fee information for
// each path. During this process, some paths may be discarded if they
// aren't able to support the total satoshis flow once fees have been
// factored in.
sourceVertex := Vertex(r.selfNode.PubKeyBytes)
validRoutes, err := pathsToFeeSortedRoutes(
sourceVertex, shortestPaths, finalCLTVDelta, amt, feeLimit,
uint32(currentHeight),
)
if err != nil {
return nil, err
}
go log.Tracef("Obtained %v paths sending %v to %x: %v", len(validRoutes),
amt, dest, newLogClosure(func() string {
return spew.Sdump(validRoutes)
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.
2016-12-27 08:20:26 +03:00
}),
)
// Populate the cache with this set of fresh routes so we can reuse
// them in the future.
r.routeCacheMtx.Lock()
r.routeCache[rt] = validRoutes
r.routeCacheMtx.Unlock()
return validRoutes, 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.
2016-12-27 08:20:26 +03:00
}
// generateSphinxPacket generates then encodes a sphinx packet which encodes
// the onion route specified by the passed layer 3 route. The blob returned
// from this function can immediately be included within an HTLC add packet to
// be sent to the first hop within the route.
func generateSphinxPacket(route *Route, paymentHash []byte) ([]byte,
*sphinx.Circuit, error) {
// As a sanity check, we'll ensure that the set of hops has been
// properly filled in, otherwise, we won't actually be able to
// construct a route.
if len(route.Hops) == 0 {
return nil, nil, ErrNoRouteHopsProvided
}
// First obtain all the public keys along the route which are contained
// in each hop.
nodes := make([]*btcec.PublicKey, len(route.Hops))
for i, hop := range route.Hops {
pub, err := btcec.ParsePubKey(hop.PubKeyBytes[:],
btcec.S256())
if err != nil {
return nil, nil, err
}
nodes[i] = pub
}
// Next we generate the per-hop payload which gives each node within
// the route the necessary information (fees, CLTV value, etc) to
// properly forward the payment.
hopPayloads := route.ToHopPayloads()
log.Tracef("Constructed per-hop payloads for payment_hash=%x: %v",
paymentHash[:], newLogClosure(func() string {
return spew.Sdump(hopPayloads)
}),
)
sessionKey, err := btcec.NewPrivateKey(btcec.S256())
if err != nil {
return nil, nil, err
}
// Next generate the onion routing packet which allows us to perform
// privacy preserving source routing across the network.
sphinxPacket, err := sphinx.NewOnionPacket(
nodes, sessionKey, hopPayloads, paymentHash,
)
if err != nil {
return nil, nil, err
}
// Finally, encode Sphinx packet using its wire representation to be
// included within the HTLC add packet.
var onionBlob bytes.Buffer
if err := sphinxPacket.Encode(&onionBlob); err != nil {
return nil, nil, err
}
log.Tracef("Generated sphinx packet: %v",
newLogClosure(func() string {
// We unset the internal curve here in order to keep
// the logs from getting noisy.
sphinxPacket.EphemeralKey.Curve = nil
return spew.Sdump(sphinxPacket)
}),
)
return onionBlob.Bytes(), &sphinx.Circuit{
SessionKey: sessionKey,
PaymentPath: nodes,
}, nil
}
// LightningPayment describes a payment to be sent through the network to the
// final destination.
type LightningPayment struct {
// Target is the node in which the payment should be routed towards.
Target *btcec.PublicKey
// Amount is the value of the payment to send through the network in
// milli-satoshis.
Amount lnwire.MilliSatoshi
// FeeLimit is the maximum fee in millisatoshis that the payment should
// accept when sending it through the network. The payment will fail
// if there isn't a route with lower fees than this limit.
FeeLimit lnwire.MilliSatoshi
// PaymentHash is the r-hash value to use within the HTLC extended to
// the first hop.
PaymentHash [32]byte
// FinalCLTVDelta is the CTLV expiry delta to use for the _final_ hop
// in the route. This means that the final hop will have a CLTV delta
// of at least: currentHeight + FinalCLTVDelta. If this value is
// unspecified, then a default value of DefaultFinalCLTVDelta will be
// used.
FinalCLTVDelta *uint16
// PayAttemptTimeout is a timeout value that we'll use to determine
// when we should should abandon the payment attempt after consecutive
// payment failure. This prevents us from attempting to send a payment
// indefinitely.
PayAttemptTimeout time.Duration
// RouteHints represents the different routing hints that can be used to
// assist a payment in reaching its destination successfully. These
// hints will act as intermediate hops along the route.
//
// NOTE: This is optional unless required by the payment. When providing
// multiple routes, ensure the hop hints within each route are chained
// together and sorted in forward order in order to reach the
// destination successfully.
RouteHints [][]HopHint
// TODO(roasbeef): add e2e message?
}
// SendPayment attempts to send a payment as described within the passed
// LightningPayment. This function is blocking and will return either: when the
// payment is successful, or all candidates routes have been attempted and
// resulted in a failed payment. If the payment succeeds, then a non-nil Route
// will be returned which describes the path the successful payment traversed
// within the network to reach the destination. Additionally, the payment
// preimage will also be returned.
func (r *ChannelRouter) SendPayment(payment *LightningPayment) ([32]byte, *Route, error) {
2018-01-30 11:03:29 +03:00
// Before starting the HTLC routing attempt, we'll create a fresh
// payment session which will report our errors back to mission
// control.
paySession, err := r.missionControl.NewPaymentSession(
payment.RouteHints, payment.Target,
)
if err != nil {
return [32]byte{}, nil, err
}
return r.sendPayment(payment, paySession)
}
// SendToRoute attempts to send a payment as described within the passed
// LightningPayment through the provided routes. This function is blocking
// and will return either: when the payment is successful, or all routes
// have been attempted and resulted in a failed payment. If the payment
// succeeds, then a non-nil Route will be returned which describes the
// path the successful payment traversed within the network to reach the
// destination. Additionally, the payment preimage will also be returned.
func (r *ChannelRouter) SendToRoute(routes []*Route,
payment *LightningPayment) ([32]byte, *Route, error) {
paySession := r.missionControl.NewPaymentSessionFromRoutes(
routes,
)
2018-01-30 11:03:29 +03:00
return r.sendPayment(payment, paySession)
}
// sendPayment attempts to send a payment as described within the passed
// LightningPayment. This function is blocking and will return either: when the
// payment is successful, or all candidates routes have been attempted and
// resulted in a failed payment. If the payment succeeds, then a non-nil Route
// will be returned which describes the path the successful payment traversed
// within the network to reach the destination. Additionally, the payment
// preimage will also be returned.
func (r *ChannelRouter) sendPayment(payment *LightningPayment,
paySession *paymentSession) ([32]byte, *Route, error) {
2018-01-30 11:03:29 +03:00
log.Tracef("Dispatching route for lightning payment: %v",
newLogClosure(func() string {
// Remove the public key curve parameters when logging
// the route to prevent spamming the logs.
if payment.Target != nil {
payment.Target.Curve = nil
}
for _, routeHint := range payment.RouteHints {
for _, hopHint := range routeHint {
hopHint.NodeID.Curve = nil
}
}
return spew.Sdump(payment)
}),
)
var (
preImage [32]byte
sendError error
)
// We'll also fetch the current block height so we can properly
// calculate the required HTLC time locks within the route.
_, currentHeight, err := r.cfg.Chain.GetBestBlock()
if err != nil {
2018-01-30 11:03:29 +03:00
return [32]byte{}, nil, err
}
var finalCLTVDelta uint16
if payment.FinalCLTVDelta == nil {
finalCLTVDelta = DefaultFinalCLTVDelta
} else {
finalCLTVDelta = *payment.FinalCLTVDelta
}
var payAttemptTimeout time.Duration
if payment.PayAttemptTimeout == time.Duration(0) {
payAttemptTimeout = defaultPayAttemptTimeout
} else {
payAttemptTimeout = payment.PayAttemptTimeout
}
timeoutChan := time.After(payAttemptTimeout)
// We'll continue until either our payment succeeds, or we encounter a
// critical error during path finding.
for {
// Before we attempt this next payment, we'll check to see if
// either we've gone past the payment attempt timeout, or the
// router is exiting. In either case, we'll stop this payment
// attempt short.
select {
case <-timeoutChan:
errStr := fmt.Sprintf("payment attempt not completed "+
"before timeout of %v", payAttemptTimeout)
return preImage, nil, newErr(
ErrPaymentAttemptTimeout, errStr,
)
case <-r.quit:
return preImage, nil, fmt.Errorf("router shutting down")
default:
// Fall through if we haven't hit our time limit, or
// are expiring.
}
route, err := paySession.RequestRoute(
payment, uint32(currentHeight), finalCLTVDelta,
)
if err != nil {
// If we're unable to successfully make a payment using
// any of the routes we've found, then return an error.
if sendError != nil {
return [32]byte{}, nil, fmt.Errorf("unable to "+
"route payment to destination: %v",
sendError)
}
return preImage, nil, err
}
log.Tracef("Attempting to send payment %x, using route: %v",
payment.PaymentHash, newLogClosure(func() string {
return spew.Sdump(route)
}),
)
// Generate the raw encoded sphinx packet to be included along
// with the htlcAdd message that we send directly to the
// switch.
onionBlob, circuit, err := generateSphinxPacket(
route, payment.PaymentHash[:],
)
if err != nil {
return preImage, nil, err
}
// Craft an HTLC packet to send to the layer 2 switch. The
// metadata within this packet will be used to route the
// payment through the network, starting with the first-hop.
htlcAdd := &lnwire.UpdateAddHTLC{
Amount: route.TotalAmount,
Expiry: route.TotalTimeLock,
PaymentHash: payment.PaymentHash,
}
copy(htlcAdd.OnionBlob[:], onionBlob)
// Attempt to send this payment through the network to complete
// the payment. If this attempt fails, then we'll continue on
// to the next available route.
firstHop := lnwire.NewShortChanIDFromInt(
route.Hops[0].ChannelID,
)
preImage, sendError = r.cfg.SendToSwitch(
firstHop, htlcAdd, circuit,
)
if sendError != nil {
// An error occurred when attempting to send the
// payment, depending on the error type, we'll either
// continue to send using alternative routes, or simply
// terminate this attempt.
log.Errorf("Attempt to send payment %x failed: %v",
payment.PaymentHash, sendError)
fErr, ok := sendError.(*htlcswitch.ForwardingError)
if !ok {
return preImage, nil, sendError
}
errSource := fErr.ErrorSource
errVertex := NewVertex(errSource)
log.Tracef("node=%x reported failure when sending "+
"htlc=%x", errVertex, payment.PaymentHash[:])
// Always determine chan id ourselves, because a channel
// update with id may not be available.
failedEdge, err := getFailedEdge(route, errVertex)
if err != nil {
return preImage, nil, err
}
// processChannelUpdateAndRetry is a closure that
// handles a failure message containing a channel
// update. This function always tries to apply the
// channel update and passes on the result to the
// payment session to adjust its view on the reliability
// of the network.
//
// As channel id, the locally determined channel id is
// used. It does not rely on the channel id that is part
// of the channel update message, because the remote
// node may lie to us or the update may be corrupt.
processChannelUpdateAndRetry := func(
update *lnwire.ChannelUpdate,
pubKey *btcec.PublicKey) {
// Try to apply the channel update.
updateOk := r.applyChannelUpdate(update, pubKey)
// If the update could not be applied, prune the
// edge. There is no reason to continue trying
// this channel.
//
// TODO: Could even prune the node completely?
// Or is there a valid reason for the channel
// update to fail?
if !updateOk {
paySession.ReportEdgeFailure(
failedEdge,
)
}
paySession.ReportEdgePolicyFailure(
NewVertex(errSource), failedEdge,
)
}
switch onionErr := fErr.FailureMessage.(type) {
// If the end destination didn't know the payment
// hash or we sent the wrong payment amount to the
// destination, then we'll terminate immediately.
case *lnwire.FailUnknownPaymentHash:
return preImage, nil, sendError
// If we sent the wrong amount to the destination, then
// we'll exit early.
case *lnwire.FailIncorrectPaymentAmount:
return preImage, nil, sendError
// If the time-lock that was extended to the final node
// was incorrect, then we can't proceed.
case *lnwire.FailFinalIncorrectCltvExpiry:
return preImage, nil, sendError
// If we crafted an invalid onion payload for the final
// node, then we'll exit early.
case *lnwire.FailFinalIncorrectHtlcAmount:
return preImage, nil, sendError
// Similarly, if the HTLC expiry that we extended to
// the final hop expires too soon, then will fail the
// payment.
//
// TODO(roasbeef): can happen to to race condition, try
// again with recent block height
case *lnwire.FailFinalExpiryTooSoon:
return preImage, nil, sendError
// If we erroneously attempted to cross a chain border,
// then we'll cancel the payment.
case *lnwire.FailInvalidRealm:
return preImage, nil, sendError
// If we get a notice that the expiry was too soon for
// an intermediate node, then we'll prune out the node
// that sent us this error, as it doesn't now what the
// correct block height is.
case *lnwire.FailExpiryTooSoon:
r.applyChannelUpdate(&onionErr.Update, errSource)
paySession.ReportVertexFailure(errVertex)
continue
// If we hit an instance of onion payload corruption or
// an invalid version, then we'll exit early as this
// shouldn't happen in the typical case.
case *lnwire.FailInvalidOnionVersion:
return preImage, nil, sendError
case *lnwire.FailInvalidOnionHmac:
return preImage, nil, sendError
case *lnwire.FailInvalidOnionKey:
return preImage, nil, sendError
// If we get a failure due to violating the minimum
// amount, we'll apply the new minimum amount and retry
// routing.
case *lnwire.FailAmountBelowMinimum:
processChannelUpdateAndRetry(
&onionErr.Update, errSource,
)
continue
// If we get a failure due to a fee, we'll apply the
// new fee update, and retry our attempt using the
// newly updated fees.
case *lnwire.FailFeeInsufficient:
processChannelUpdateAndRetry(
&onionErr.Update, errSource,
)
continue
// If we get the failure for an intermediate node that
// disagrees with our time lock values, then we'll
// apply the new delta value and try it once more.
case *lnwire.FailIncorrectCltvExpiry:
processChannelUpdateAndRetry(
&onionErr.Update, errSource,
)
continue
// The outgoing channel that this node was meant to
// forward one is currently disabled, so we'll apply
// the update and continue.
case *lnwire.FailChannelDisabled:
r.applyChannelUpdate(&onionErr.Update, errSource)
paySession.ReportEdgeFailure(failedEdge)
continue
// It's likely that the outgoing channel didn't have
// sufficient capacity, so we'll prune this edge for
// now, and continue onwards with our path finding.
case *lnwire.FailTemporaryChannelFailure:
r.applyChannelUpdate(onionErr.Update, errSource)
paySession.ReportEdgeFailure(failedEdge)
continue
// If the send fail due to a node not having the
// required features, then we'll note this error and
// continue.
case *lnwire.FailRequiredNodeFeatureMissing:
paySession.ReportVertexFailure(errVertex)
continue
// If the send fail due to a node not having the
// required features, then we'll note this error and
// continue.
case *lnwire.FailRequiredChannelFeatureMissing:
paySession.ReportVertexFailure(errVertex)
continue
// If the next hop in the route wasn't known or
// offline, we'll only the channel which we attempted
// to route over. This is conservative, and it can
// handle faulty channels between nodes properly.
// Additionally, this guards against routing nodes
// returning errors in order to attempt to black list
// another node.
case *lnwire.FailUnknownNextPeer:
paySession.ReportEdgeFailure(failedEdge)
continue
// If the node wasn't able to forward for which ever
// reason, then we'll note this and continue with the
// routes.
case *lnwire.FailTemporaryNodeFailure:
paySession.ReportVertexFailure(errVertex)
continue
case *lnwire.FailPermanentNodeFailure:
paySession.ReportVertexFailure(errVertex)
continue
// If we crafted a route that contains a too long time
// lock for an intermediate node, we'll prune the node.
// As there currently is no way of knowing that node's
// maximum acceptable cltv, we cannot take this
// constraint into account during routing.
//
// TODO(joostjager): Record the rejected cltv and use
// that as a hint during future path finding through
// that node.
case *lnwire.FailExpiryTooFar:
paySession.ReportVertexFailure(errVertex)
continue
// If we get a permanent channel or node failure, then
// we'll prune the channel in both directions and
// continue with the rest of the routes.
case *lnwire.FailPermanentChannelFailure:
paySession.ReportEdgeFailure(&edgeLocator{
channelID: failedEdge.channelID,
direction: 0,
})
paySession.ReportEdgeFailure(&edgeLocator{
channelID: failedEdge.channelID,
direction: 1,
})
continue
default:
return preImage, nil, sendError
}
}
return preImage, route, nil
}
}
// getFailedEdge tries to locate the failing channel given a route and the
// pubkey of the node that sent the error. It will assume that the error is
// associated with the outgoing channel of the error node.
func getFailedEdge(route *Route, errSource Vertex) (
*edgeLocator, error) {
hopCount := len(route.Hops)
fromNode := route.SourcePubKey
for i, hop := range route.Hops {
toNode := hop.PubKeyBytes
// Determine if we have a failure from the final hop.
//
// TODO(joostjager): In this case, certain types of errors are
// not expected. For example FailUnknownNextPeer. This could be
// a reason to prune the node?
finalHopFailing := i == hopCount-1 && errSource == toNode
// As this error indicates that the target channel was unable to
// carry this HTLC (for w/e reason), we'll return the _outgoing_
// channel that the source of the error was meant to pass the
// HTLC along to.
//
// If the errSource is the final hop, we assume that the failing
// channel is the incoming channel.
if errSource == fromNode || finalHopFailing {
return newEdgeLocatorByPubkeys(
hop.ChannelID,
&fromNode,
&toNode,
), nil
}
fromNode = toNode
}
return nil, fmt.Errorf("cannot find error source node in route")
}
// applyChannelUpdate validates a channel update and if valid, applies it to the
// database. It returns a bool indicating whether the updates was successful.
func (r *ChannelRouter) applyChannelUpdate(msg *lnwire.ChannelUpdate,
pubKey *btcec.PublicKey) bool {
// If we get passed a nil channel update (as it's optional with some
// onion errors), then we'll exit early with a success result.
if msg == nil {
return true
}
ch, _, _, err := r.GetChannelByID(msg.ShortChannelID)
if err != nil {
log.Errorf("Unable to retrieve channel by id: %v", err)
return false
}
if err := ValidateChannelUpdateAnn(pubKey, ch.Capacity, msg); err != nil {
log.Errorf("Unable to validate channel update: %v", err)
return false
}
err = r.UpdateEdge(&channeldb.ChannelEdgePolicy{
SigBytes: msg.Signature.ToSignatureBytes(),
ChannelID: msg.ShortChannelID.ToUint64(),
LastUpdate: time.Unix(int64(msg.Timestamp), 0),
MessageFlags: msg.MessageFlags,
ChannelFlags: msg.ChannelFlags,
TimeLockDelta: msg.TimeLockDelta,
MinHTLC: msg.HtlcMinimumMsat,
MaxHTLC: msg.HtlcMaximumMsat,
FeeBaseMSat: lnwire.MilliSatoshi(msg.BaseFee),
FeeProportionalMillionths: lnwire.MilliSatoshi(msg.FeeRate),
})
if err != nil && !IsError(err, ErrIgnored, ErrOutdated) {
log.Errorf("Unable to apply channel update: %v", err)
return false
}
return true
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.
2016-12-27 08:20:26 +03:00
}
// AddNode is used to add information about a node to the router database. If
// the node with this pubkey is not present in an existing channel, it will
// be ignored.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) AddNode(node *channeldb.LightningNode) error {
rMsg := &routingMsg{
msg: node,
err: make(chan error, 1),
}
select {
case r.networkUpdates <- rMsg:
select {
case err := <-rMsg.err:
return err
case <-r.quit:
return errors.New("router has been shut down")
}
case <-r.quit:
return errors.New("router has been shut down")
}
}
// AddEdge is used to add edge/channel to the topology of the router, after all
// information about channel will be gathered this edge/channel might be used
// in construction of payment path.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) AddEdge(edge *channeldb.ChannelEdgeInfo) error {
rMsg := &routingMsg{
msg: edge,
err: make(chan error, 1),
}
select {
case r.networkUpdates <- rMsg:
select {
case err := <-rMsg.err:
return err
case <-r.quit:
return errors.New("router has been shut down")
}
case <-r.quit:
return errors.New("router has been shut down")
}
}
// UpdateEdge is used to update edge information, without this message edge
// considered as not fully constructed.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) UpdateEdge(update *channeldb.ChannelEdgePolicy) error {
rMsg := &routingMsg{
msg: update,
err: make(chan error, 1),
}
select {
case r.networkUpdates <- rMsg:
select {
case err := <-rMsg.err:
return err
case <-r.quit:
return errors.New("router has been shut down")
}
case <-r.quit:
return errors.New("router has been shut down")
}
}
// CurrentBlockHeight returns the block height from POV of the router subsystem.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) CurrentBlockHeight() (uint32, error) {
_, height, err := r.cfg.Chain.GetBestBlock()
return uint32(height), err
}
// GetChannelByID return the channel by the channel id.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) GetChannelByID(chanID lnwire.ShortChannelID) (
*channeldb.ChannelEdgeInfo,
*channeldb.ChannelEdgePolicy,
*channeldb.ChannelEdgePolicy, error) {
return r.cfg.Graph.FetchChannelEdgesByID(chanID.ToUint64())
}
// FetchLightningNode attempts to look up a target node by its identity public
// key. channeldb.ErrGraphNodeNotFound is returned if the node doesn't exist
// within the graph.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) FetchLightningNode(node Vertex) (*channeldb.LightningNode, error) {
pubKey, err := btcec.ParsePubKey(node[:], btcec.S256())
if err != nil {
return nil, fmt.Errorf("unable to parse raw public key: %v", err)
}
return r.cfg.Graph.FetchLightningNode(pubKey)
}
// ForEachNode is used to iterate over every node in router topology.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) ForEachNode(cb func(*channeldb.LightningNode) error) error {
return r.cfg.Graph.ForEachNode(nil, func(_ *bbolt.Tx, n *channeldb.LightningNode) error {
return cb(n)
})
}
// ForAllOutgoingChannels is used to iterate over all outgoing channels owned by
// the router.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) ForAllOutgoingChannels(cb func(*channeldb.ChannelEdgeInfo,
*channeldb.ChannelEdgePolicy) error) error {
return r.selfNode.ForEachChannel(nil, func(_ *bbolt.Tx, c *channeldb.ChannelEdgeInfo,
e, _ *channeldb.ChannelEdgePolicy) error {
if e == nil {
return fmt.Errorf("Channel from self node has no policy")
}
return cb(c, e)
})
}
// ForEachChannel is used to iterate over every known edge (channel) within our
// view of the channel graph.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) ForEachChannel(cb func(chanInfo *channeldb.ChannelEdgeInfo,
e1, e2 *channeldb.ChannelEdgePolicy) error) error {
return r.cfg.Graph.ForEachChannel(cb)
}
// AddProof updates the channel edge info with proof which is needed to
// properly announce the edge to the rest of the network.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) AddProof(chanID lnwire.ShortChannelID,
proof *channeldb.ChannelAuthProof) error {
info, _, _, err := r.cfg.Graph.FetchChannelEdgesByID(chanID.ToUint64())
if err != nil {
return err
}
info.AuthProof = proof
return r.cfg.Graph.UpdateChannelEdge(info)
}
// IsStaleNode returns true if the graph source has a node announcement for the
// target node with a more recent timestamp.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsStaleNode(node Vertex, timestamp time.Time) bool {
// If our attempt to assert that the node announcement is fresh fails,
// then we know that this is actually a stale announcement.
return r.assertNodeAnnFreshness(node, timestamp) != nil
}
// IsPublicNode determines whether the given vertex is seen as a public node in
// the graph from the graph's source node's point of view.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsPublicNode(node Vertex) (bool, error) {
return r.cfg.Graph.IsPublicNode(node)
}
// IsKnownEdge returns true if the graph source already knows of the passed
// channel ID.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsKnownEdge(chanID lnwire.ShortChannelID) bool {
_, _, exists, _ := r.cfg.Graph.HasChannelEdge(chanID.ToUint64())
return exists
}
// IsStaleEdgePolicy returns true if the graph soruce has a channel edge for
// the passed channel ID (and flags) that have a more recent timestamp.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsStaleEdgePolicy(chanID lnwire.ShortChannelID,
timestamp time.Time, flags lnwire.ChanUpdateChanFlags) bool {
edge1Timestamp, edge2Timestamp, exists, err := r.cfg.Graph.HasChannelEdge(
chanID.ToUint64(),
)
if err != nil {
return false
}
// If we don't know of the edge, then it means it's fresh (thus not
// stale).
if !exists {
return false
}
// As edges are directional edge node has a unique policy for the
// direction of the edge they control. Therefore we first check if we
// already have the most up to date information for that edge. If so,
// then we can exit early.
switch {
// A flag set of 0 indicates this is an announcement for the "first"
// node in the channel.
case flags&lnwire.ChanUpdateDirection == 0:
return !edge1Timestamp.Before(timestamp)
// Similarly, a flag set of 1 indicates this is an announcement for the
// "second" node in the channel.
case flags&lnwire.ChanUpdateDirection == 1:
return !edge2Timestamp.Before(timestamp)
}
return false
}