electrum

Electrum Bitcoin wallet
git clone https://git.parazyd.org/electrum
Log | Files | Refs | Submodules

lnrater.py (11416B)


      1 # Copyright (C) 2020 The Electrum developers
      2 # Distributed under the MIT software license, see the accompanying
      3 # file LICENCE or http://www.opensource.org/licenses/mit-license.php
      4 """
      5 lnrater.py contains Lightning Network node rating functionality.
      6 """
      7 
      8 import asyncio
      9 from collections import defaultdict
     10 from pprint import pformat
     11 from random import choices
     12 from statistics import mean, median, stdev
     13 from typing import TYPE_CHECKING, Dict, NamedTuple, Tuple, List, Optional
     14 import sys
     15 import time
     16 
     17 if sys.version_info[:2] >= (3, 7):
     18     from asyncio import get_running_loop
     19 else:
     20     from asyncio import _get_running_loop as get_running_loop  # noqa: F401
     21 
     22 from .logging import Logger
     23 from .util import profiler
     24 from .lnrouter import fee_for_edge_msat
     25 from .lnutil import LnFeatures, ln_compare_features, IncompatibleLightningFeatures
     26 
     27 if TYPE_CHECKING:
     28     from .network import Network
     29     from .channel_db import Policy, NodeInfo
     30     from .lnchannel import ShortChannelID
     31     from .lnworker import LNWallet
     32 
     33 
     34 MONTH_IN_BLOCKS = 6 * 24 * 30
     35 # the scores are only updated after this time interval
     36 RATER_UPDATE_TIME_SEC = 10 * 60
     37 # amount used for calculating an effective relative fee
     38 FEE_AMOUNT_MSAT = 100_000_000
     39 
     40 # define some numbers for minimal requirements of good nodes
     41 # exclude nodes with less number of channels
     42 EXCLUDE_NUM_CHANNELS = 15
     43 # exclude nodes with less mean capacity
     44 EXCLUDE_MEAN_CAPACITY_MSAT = 1_000_000_000
     45 # exclude nodes which are young
     46 EXCLUDE_NODE_AGE = 2 * MONTH_IN_BLOCKS
     47 # exclude nodes which have young mean channel age
     48 EXCLUDE_MEAN_CHANNEL_AGE = EXCLUDE_NODE_AGE
     49 # exclude nodes which charge a high fee
     50 EXCLUCE_EFFECTIVE_FEE_RATE = 0.001500
     51 # exclude nodes whose last channel open was a long time ago
     52 EXCLUDE_BLOCKS_LAST_CHANNEL = 3 * MONTH_IN_BLOCKS
     53 
     54 
     55 class NodeStats(NamedTuple):
     56     number_channels: int
     57     # capacity related
     58     total_capacity_msat: int
     59     median_capacity_msat: float
     60     mean_capacity_msat: float
     61     # block height related
     62     node_age_block_height: int
     63     mean_channel_age_block_height: float
     64     blocks_since_last_channel: int
     65     # fees
     66     mean_fee_rate: float
     67 
     68 
     69 def weighted_sum(numbers: List[float], weights: List[float]) -> float:
     70     running_sum = 0.0
     71     for n, w in zip(numbers, weights):
     72         running_sum += n * w
     73     return running_sum/sum(weights)
     74 
     75 
     76 class LNRater(Logger):
     77     def __init__(self, lnworker: 'LNWallet', network: 'Network'):
     78         """LNRater can be used to suggest nodes to open up channels with.
     79 
     80         The graph is analyzed and some heuristics are applied to sort out nodes
     81         that are deemed to be bad routers or unmaintained.
     82         """
     83         Logger.__init__(self)
     84         self.lnworker = lnworker
     85         self.network = network
     86 
     87         self._node_stats: Dict[bytes, NodeStats] = {}  # node_id -> NodeStats
     88         self._node_ratings: Dict[bytes, float] = {}  # node_id -> float
     89         self._policies_by_nodes: Dict[bytes, List[Tuple[ShortChannelID, Policy]]] = defaultdict(list)  # node_id -> (short_channel_id, policy)
     90         self._last_analyzed = 0  # timestamp
     91         self._last_progress_percent = 0
     92 
     93     def maybe_analyze_graph(self):
     94         loop = asyncio.get_event_loop()
     95         fut = asyncio.run_coroutine_threadsafe(self._maybe_analyze_graph(), loop)
     96         fut.result()
     97 
     98     def analyze_graph(self):
     99         """Forces a graph analysis, e.g., due to external triggers like
    100         the graph info reaching 50%."""
    101         loop = asyncio.get_event_loop()
    102         fut = asyncio.run_coroutine_threadsafe(self._analyze_graph(), loop)
    103         fut.result()
    104 
    105     async def _maybe_analyze_graph(self):
    106         """Analyzes the graph when in early sync stage (>30%) or when caching
    107         time expires."""
    108         # gather information about graph sync status
    109         current_channels, total, progress_percent = self.network.lngossip.get_sync_progress_estimate()
    110 
    111         # gossip sync progress state could be None when not started, but channel
    112         # db already knows something about the graph, which is why we allow to
    113         # evaluate the graph early
    114         if progress_percent is not None or self.network.channel_db.num_nodes > 500:
    115             progress_percent = progress_percent or 0  # convert None to 0
    116             now = time.time()
    117             # graph should have changed significantly during the sync progress
    118             # or last analysis was a long time ago
    119             if (30 <= progress_percent and progress_percent - self._last_progress_percent >= 10 or
    120                     self._last_analyzed + RATER_UPDATE_TIME_SEC < now):
    121                 await self._analyze_graph()
    122                 self._last_progress_percent = progress_percent
    123                 self._last_analyzed = now
    124 
    125     async def _analyze_graph(self):
    126         await self.network.channel_db.data_loaded.wait()
    127         self._collect_policies_by_node()
    128         loop = get_running_loop()
    129         # the analysis is run in an executor because it's costly
    130         await loop.run_in_executor(None, self._collect_purged_stats)
    131         self._rate_nodes()
    132         now = time.time()
    133         self._last_analyzed = now
    134 
    135     def _collect_policies_by_node(self):
    136         policies = self.network.channel_db.get_node_policies()
    137         for pv, p in policies.items():
    138             # append tuples of ShortChannelID and Policy
    139             self._policies_by_nodes[pv[0]].append((pv[1], p))
    140 
    141     @profiler
    142     def _collect_purged_stats(self):
    143         """Traverses through the graph and sorts out nodes."""
    144         current_height = self.network.get_local_height()
    145         node_infos = self.network.channel_db.get_node_infos()
    146 
    147         for n, channel_policies in self._policies_by_nodes.items():
    148             try:
    149                 # use policies synonymously to channels
    150                 num_channels = len(channel_policies)
    151 
    152                 # save some time for nodes we are not interested in:
    153                 if num_channels < EXCLUDE_NUM_CHANNELS:
    154                     continue
    155 
    156                 # analyze block heights
    157                 block_heights = [p[0].block_height for p in channel_policies]
    158                 node_age_bh = current_height - min(block_heights)
    159                 if node_age_bh < EXCLUDE_NODE_AGE:
    160                     continue
    161                 mean_channel_age_bh = current_height - mean(block_heights)
    162                 if mean_channel_age_bh < EXCLUDE_MEAN_CHANNEL_AGE:
    163                     continue
    164                 blocks_since_last_channel = current_height - max(block_heights)
    165                 if blocks_since_last_channel > EXCLUDE_BLOCKS_LAST_CHANNEL:
    166                     continue
    167 
    168                 # analyze capacities
    169                 capacities = [p[1].htlc_maximum_msat for p in channel_policies]
    170                 if None in capacities:
    171                     continue
    172                 total_capacity = sum(capacities)
    173 
    174                 mean_capacity = total_capacity / num_channels if num_channels else 0
    175                 if mean_capacity < EXCLUDE_MEAN_CAPACITY_MSAT:
    176                     continue
    177                 median_capacity = median(capacities)
    178 
    179                 # analyze fees
    180                 effective_fee_rates = [fee_for_edge_msat(
    181                     FEE_AMOUNT_MSAT,
    182                     p[1].fee_base_msat,
    183                     p[1].fee_proportional_millionths) / FEE_AMOUNT_MSAT for p in channel_policies]
    184                 mean_fees_rate = mean(effective_fee_rates)
    185                 if mean_fees_rate > EXCLUCE_EFFECTIVE_FEE_RATE:
    186                     continue
    187 
    188                 self._node_stats[n] = NodeStats(
    189                     number_channels=num_channels,
    190                     total_capacity_msat=total_capacity,
    191                     median_capacity_msat=median_capacity,
    192                     mean_capacity_msat=mean_capacity,
    193                     node_age_block_height=node_age_bh,
    194                     mean_channel_age_block_height=mean_channel_age_bh,
    195                     blocks_since_last_channel=blocks_since_last_channel,
    196                     mean_fee_rate=mean_fees_rate
    197                 )
    198 
    199             except Exception as e:
    200                 self.logger.exception("Could not use channel policies for "
    201                                       "calculating statistics.")
    202                 self.logger.debug(pformat(channel_policies))
    203                 continue
    204 
    205         self.logger.info(f"node statistics done, calculated statistics"
    206                          f"for {len(self._node_stats)} nodes")
    207 
    208     def _rate_nodes(self):
    209         """Rate nodes by collected statistics."""
    210 
    211         max_capacity = 0
    212         max_num_chan = 0
    213         min_fee_rate = float('inf')
    214         for stats in self._node_stats.values():
    215             max_capacity = max(max_capacity, stats.total_capacity_msat)
    216             max_num_chan = max(max_num_chan, stats.number_channels)
    217             min_fee_rate = min(min_fee_rate, stats.mean_fee_rate)
    218 
    219         for n, stats in self._node_stats.items():
    220             heuristics = []
    221             heuristics_weights = []
    222 
    223             # Construct an average score which leads to recommendation of nodes
    224             # with low fees, large capacity and reasonable number of channels.
    225             # This is somewhat akin to preferential attachment, but low fee
    226             # nodes are more favored. Here we make a compromise between user
    227             # comfort and decentralization, tending towards user comfort.
    228 
    229             # number of channels
    230             heuristics.append(stats.number_channels / max_num_chan)
    231             heuristics_weights.append(0.2)
    232             # total capacity
    233             heuristics.append(stats.total_capacity_msat / max_capacity)
    234             heuristics_weights.append(0.8)
    235             # inverse fees
    236             fees = min(1E-6, min_fee_rate) / max(1E-10, stats.mean_fee_rate)
    237             heuristics.append(fees)
    238             heuristics_weights.append(1.0)
    239 
    240             self._node_ratings[n] = weighted_sum(heuristics, heuristics_weights)
    241 
    242     def suggest_node_channel_open(self) -> Tuple[bytes, NodeStats]:
    243         node_keys = list(self._node_stats.keys())
    244         node_ratings = list(self._node_ratings.values())
    245         channel_peers = self.lnworker.channel_peers()
    246         node_info: Optional["NodeInfo"] = None
    247 
    248         while True:
    249             # randomly pick nodes weighted by node_rating
    250             pk = choices(node_keys, weights=node_ratings, k=1)[0]
    251             # node should have compatible features
    252             node_info = self.network.channel_db.get_node_infos().get(pk, None)
    253             peer_features = LnFeatures(node_info.features)
    254             try:
    255                 ln_compare_features(self.lnworker.features, peer_features)
    256             except IncompatibleLightningFeatures as e:
    257                 self.logger.info("suggested node is incompatible")
    258                 continue
    259 
    260             # don't want to connect to nodes we are already connected to
    261             if pk not in channel_peers:
    262                 break
    263 
    264         alias = node_info.alias if node_info else 'unknown node alias'
    265         self.logger.info(
    266             f"node rating for {alias}:\n"
    267             f"{pformat(self._node_stats[pk])} (score {self._node_ratings[pk]})")
    268 
    269         return pk, self._node_stats[pk]
    270 
    271     def suggest_peer(self) -> Optional[bytes]:
    272         """Suggests a LN node to open a channel with.
    273         Returns a node ID (pubkey).
    274         """
    275         self.maybe_analyze_graph()
    276         if self._node_ratings:
    277             return self.suggest_node_channel_open()[0]
    278         else:
    279             return None