autopilot/interface+agent: remove Select

This commit is contained in:
Johan T. Halseth 2018-11-22 23:18:09 +01:00
parent b3d315298c
commit 6130189d95
No known key found for this signature in database
GPG Key ID: 15BAADA29DA20D26
3 changed files with 0 additions and 198 deletions

@ -66,12 +66,6 @@ type directiveArg struct {
nodes map[NodeID]struct{}
}
func (m *mockHeuristic) Select(self *btcec.PublicKey, graph ChannelGraph,
amtToUse btcutil.Amount, numChans uint32,
skipChans map[NodeID]struct{}) ([]AttachmentDirective, error) {
return nil, nil
}
func (m *mockHeuristic) NodeScores(g ChannelGraph, chans []Channel,
fundsAvailable btcutil.Amount, nodes map[NodeID]struct{}) (
map[NodeID]*AttachmentDirective, error) {

@ -121,17 +121,6 @@ type AttachmentHeuristic interface {
// ideal state.
NeedMoreChans(chans []Channel, balance btcutil.Amount) (btcutil.Amount, uint32, bool)
// Select is a method that given the current state of the channel
// graph, a set of nodes to ignore, and an amount of available funds,
// should return a set of attachment directives which describe which
// additional channels should be opened within the graph to push the
// heuristic back towards its equilibrium state. The numNewChans
// argument represents the additional number of channels that should be
// open.
Select(self *btcec.PublicKey, graph ChannelGraph,
amtToUse btcutil.Amount, numNewChans uint32,
skipNodes map[NodeID]struct{}) ([]AttachmentDirective, error)
// NodeScores is a method that given the current channel graph, current
// set of local channels and funds available, scores the given nodes
// according to the preference of opening a channel with them. The

@ -1,8 +1,6 @@
package autopilot
import (
"bytes"
"fmt"
prand "math/rand"
"net"
"time"
@ -72,185 +70,6 @@ func NewNodeID(pub *btcec.PublicKey) NodeID {
return n
}
// shuffleCandidates shuffles the set of candidate nodes for preferential
// attachment in order to break any ordering already enforced by the sorted
// order of the public key for each node. To shuffle the set of candidates, we
// use a version of the FisherYates shuffle algorithm.
func shuffleCandidates(candidates []Node) []Node {
shuffledNodes := make([]Node, len(candidates))
perm := prand.Perm(len(candidates))
for i, v := range perm {
shuffledNodes[v] = candidates[i]
}
return shuffledNodes
}
// Select returns a candidate set of attachment directives that should be
// executed based on the current internal state, the state of the channel
// graph, the set of nodes we should exclude, and the amount of funds
// available. The heuristic employed by this method is one that attempts to
// promote a scale-free network globally, via local attachment preferences for
// new nodes joining the network with an amount of available funds to be
// allocated to channels. Specifically, we consider the degree of each node
// (and the flow in/out of the node available via its open channels) and
// utilize the BarabásiAlbert model to drive our recommended attachment
// heuristics. If implemented globally for each new participant, this results
// in a channel graph that is scale-free and follows a power law distribution
// with k=-3.
//
// NOTE: This is a part of the AttachmentHeuristic interface.
func (p *ConstrainedPrefAttachment) Select(self *btcec.PublicKey, g ChannelGraph,
fundsAvailable btcutil.Amount, numNewChans uint32,
skipNodes map[NodeID]struct{}) ([]AttachmentDirective, error) {
// TODO(roasbeef): rename?
var directives []AttachmentDirective
if fundsAvailable < p.constraints.MinChanSize {
return directives, nil
}
selfPubBytes := self.SerializeCompressed()
// We'll continue our attachment loop until we've exhausted the current
// amount of available funds.
visited := make(map[NodeID]struct{})
for i := uint32(0); i < numNewChans; i++ {
// selectionSlice will be used to randomly select a node
// according to a power law distribution. For each connected
// edge, we'll add an instance of the node to this slice. Thus,
// for a given node, the probability that we'll attach to it
// is: k_i / sum(k_j), where k_i is the degree of the target
// node, and k_j is the degree of all other nodes i != j. This
// implements the classic BarabásiAlbert model for
// preferential attachment.
var selectionSlice []Node
// For each node, and each channel that the node has, we'll add
// an instance of that node to the selection slice above.
// This'll slice where the frequency of each node is equivalent
// to the number of channels that connect to it.
//
// TODO(roasbeef): add noise to make adversarially resistant?
if err := g.ForEachNode(func(node Node) error {
nID := NodeID(node.PubKey())
// Once a node has already been attached to, we'll
// ensure that it isn't factored into any further
// decisions within this round.
if _, ok := visited[nID]; ok {
return nil
}
// If we come across ourselves, them we'll continue in
// order to avoid attempting to make a channel with
// ourselves.
if bytes.Equal(nID[:], selfPubBytes) {
return nil
}
// Additionally, if this node is in the blacklist, then
// we'll skip it.
if _, ok := skipNodes[nID]; ok {
return nil
}
// For initial bootstrap purposes, if a node doesn't
// have any channels, then we'll ensure that it has at
// least one item in the selection slice.
//
// TODO(roasbeef): make conditional?
selectionSlice = append(selectionSlice, node)
// For each active channel the node has, we'll add an
// additional channel to the selection slice to
// increase their weight.
if err := node.ForEachChannel(func(channel ChannelEdge) error {
selectionSlice = append(selectionSlice, node)
return nil
}); err != nil {
return err
}
return nil
}); err != nil {
return nil, err
}
// If no nodes at all were accumulated, then we'll exit early
// as there are no eligible candidates.
if len(selectionSlice) == 0 {
break
}
// Given our selection slice, we'll now generate a random index
// into this slice. The node we select will be recommended by
// us to create a channel to.
candidates := shuffleCandidates(selectionSlice)
selectedIndex := prand.Int31n(int32(len(candidates)))
selectedNode := candidates[selectedIndex]
// TODO(roasbeef): cap on num channels to same participant?
// With the node selected, we'll add this (node, amount) tuple
// to out set of recommended directives.
pubBytes := selectedNode.PubKey()
nID := NodeID(pubBytes)
directives = append(directives, AttachmentDirective{
NodeID: nID,
Addrs: selectedNode.Addrs(),
})
// With the node selected, we'll add it to the set of visited
// nodes to avoid attaching to it again.
visited[nID] = struct{}{}
}
numSelectedNodes := int64(len(directives))
switch {
// If we have enough available funds to distribute the maximum channel
// size for each of the selected peers to attach to, then we'll
// allocate the maximum amount to each peer.
case int64(fundsAvailable) >= numSelectedNodes*int64(p.constraints.MaxChanSize):
for i := 0; i < int(numSelectedNodes); i++ {
directives[i].ChanAmt = p.constraints.MaxChanSize
}
return directives, nil
// Otherwise, we'll greedily allocate our funds to the channels
// successively until we run out of available funds, or can't create a
// channel above the min channel size.
case int64(fundsAvailable) < numSelectedNodes*int64(p.constraints.MaxChanSize):
i := 0
for fundsAvailable > p.constraints.MinChanSize {
// We'll attempt to allocate the max channel size
// initially. If we don't have enough funds to do this,
// then we'll allocate the remainder of the funds
// available to the channel.
delta := p.constraints.MaxChanSize
if fundsAvailable-delta < 0 {
delta = fundsAvailable
}
directives[i].ChanAmt = delta
fundsAvailable -= delta
i++
}
// We'll slice the initial set of directives to properly
// reflect the amount of funds we were able to allocate.
return directives[:i:i], nil
default:
return nil, fmt.Errorf("err")
}
}
// NodeScores is a method that given the current channel graph, current set of
// local channels and funds available, scores the given nodes according the the
// preference of opening a channel with them.