Stop decaying historical liquidity information during scoring

Because scoring is an incredibly performance-sensitive operation,
doing liquidity information decay (and especially fetching the
current time!) during scoring isn't really a great idea. Now that
we decay liquidity information in the background, we don't have any
reason to decay during scoring, and we remove the historical bucket
liquidity decaying here.

Author: TheBlueMatt

lightning/src/routing/scoring.rs

@@ -851,8 +851,6 @@ impl<G: Deref<Target = NetworkGraph<L>>, L: Deref, T: Time> ProbabilisticScorerU
 	/// Note that this writes roughly one line per channel for which we have a liquidity estimate,
 	/// which may be a substantial amount of log output.
 	pub fn debug_log_liquidity_stats(&self) {
-		let now = T::now();
-
 		let graph = self.network_graph.read_only();
 		for (scid, liq) in self.channel_liquidities.iter() {
 			if let Some(chan_debug) = graph.channels().get(scid) {
@@ -861,10 +859,8 @@ impl<G: Deref<Target = NetworkGraph<L>>, L: Deref, T: Time> ProbabilisticScorerU
 						let amt = directed_info.effective_capacity().as_msat();
 						let dir_liq = liq.as_directed(source, target, amt, self.decay_params);
 
-						let (min_buckets, max_buckets) = dir_liq.liquidity_history
-							.get_decayed_buckets(now, *dir_liq.offset_history_last_updated,
-								self.decay_params.historical_no_updates_half_life)
-							.unwrap_or(([0; 32], [0; 32]));
+						let min_buckets = &dir_liq.liquidity_history.min_liquidity_offset_history.buckets;
+						let max_buckets = &dir_liq.liquidity_history.max_liquidity_offset_history.buckets;
 
 						log_debug!(self.logger, core::concat!(
 							"Liquidity from {} to {} via {} is in the range ({}, {}).\n",
@@ -943,7 +939,7 @@ impl<G: Deref<Target = NetworkGraph<L>>, L: Deref, T: Time> ProbabilisticScorerU
 	/// in the top and bottom bucket, and roughly with similar (recent) frequency.
 	///
 	/// Because the datapoints are decayed slowly over time, values will eventually return to
-	/// `Some(([1; 32], [1; 32]))` and then to `None` once no datapoints remain.
+	/// `Some(([0; 32], [0; 32]))` or `None` if no data remains for a channel.
 	///
 	/// In order to fetch a single success probability from the buckets provided here, as used in
 	/// the scoring model, see [`Self::historical_estimated_payment_success_probability`].
@@ -957,11 +953,8 @@ impl<G: Deref<Target = NetworkGraph<L>>, L: Deref, T: Time> ProbabilisticScorerU
 					let amt = directed_info.effective_capacity().as_msat();
 					let dir_liq = liq.as_directed(source, target, amt, self.decay_params);
 
-					let (min_buckets, mut max_buckets) =
-						dir_liq.liquidity_history.get_decayed_buckets(
-							dir_liq.now, *dir_liq.offset_history_last_updated,
-							self.decay_params.historical_no_updates_half_life
-						)?;
+					let min_buckets = dir_liq.liquidity_history.min_liquidity_offset_history.buckets;
+					let mut max_buckets = dir_liq.liquidity_history.max_liquidity_offset_history.buckets;
 
 					// Note that the liquidity buckets are an offset from the edge, so we inverse
 					// the max order to get the probabilities from zero.
@@ -992,9 +985,7 @@ impl<G: Deref<Target = NetworkGraph<L>>, L: Deref, T: Time> ProbabilisticScorerU
 					let dir_liq = liq.as_directed(source, target, capacity_msat, self.decay_params);
 
 					return dir_liq.liquidity_history.calculate_success_probability_times_billion(
-						dir_liq.now, *dir_liq.offset_history_last_updated,
-						self.decay_params.historical_no_updates_half_life, &params, amount_msat,
-						capacity_msat
+						&params, amount_msat, capacity_msat
 					).map(|p| p as f64 / (1024 * 1024 * 1024) as f64);
 				}
 			}
@@ -1215,9 +1206,7 @@ impl<L: Deref<Target = u64>, BRT: Deref<Target = HistoricalBucketRangeTracker>,
 		   score_params.historical_liquidity_penalty_amount_multiplier_msat != 0 {
 			if let Some(cumulative_success_prob_times_billion) = self.liquidity_history
 				.calculate_success_probability_times_billion(
-					self.now, *self.offset_history_last_updated,
-					self.decay_params.historical_no_updates_half_life, score_params, amount_msat,
-					self.capacity_msat)
+					score_params, amount_msat, self.capacity_msat)
 			{
 				let historical_negative_log10_times_2048 = approx::negative_log10_times_2048(cumulative_success_prob_times_billion + 1, 1024 * 1024 * 1024);
 				res = res.saturating_add(Self::combined_penalty_msat(amount_msat,
@@ -2009,22 +1998,20 @@ mod bucketed_history {
 	}
 
 	impl<D: Deref<Target = HistoricalBucketRangeTracker>> HistoricalMinMaxBuckets<D> {
-		pub(super) fn get_decayed_buckets<T: Time>(&self, now: T, offset_history_last_updated: T, half_life: Duration)
-		-> Option<([u16; 32], [u16; 32])> {
-			let (_, required_decays) = self.get_total_valid_points(now, offset_history_last_updated, half_life)?;
-
-			let mut min_buckets = *self.min_liquidity_offset_history;
-			min_buckets.time_decay_data(required_decays);
-			let mut max_buckets = *self.max_liquidity_offset_history;
-			max_buckets.time_decay_data(required_decays);
-			Some((min_buckets.buckets, max_buckets.buckets))
-		}
 		#[inline]
-		pub(super) fn get_total_valid_points<T: Time>(&self, now: T, offset_history_last_updated: T, half_life: Duration)
-		-> Option<(u64, u32)> {
-			let required_decays = now.duration_since(offset_history_last_updated).as_secs()
-				.checked_div(half_life.as_secs())
-				.map_or(u32::max_value(), |decays| cmp::min(decays, u32::max_value() as u64) as u32);
+		pub(super) fn calculate_success_probability_times_billion(
+			&self, params: &ProbabilisticScoringFeeParameters, amount_msat: u64,
+			capacity_msat: u64
+		) -> Option<u64> {
+			// If historical penalties are enabled, we try to calculate a probability of success
+			// given our historical distribution of min- and max-liquidity bounds in a channel.
+			// To do so, we walk the set of historical liquidity bucket (min, max) combinations
+			// (where min_idx < max_idx, as having a minimum above our maximum is an invalid
+			// state). For each pair, we calculate the probability as if the bucket's corresponding
+			// min- and max- liquidity bounds were our current liquidity bounds and then multiply
+			// that probability by the weight of the selected buckets.
+			let payment_pos = amount_to_pos(amount_msat, capacity_msat);
+			if payment_pos >= POSITION_TICKS { return None; }
 
 			let mut total_valid_points_tracked = 0;
 			for (min_idx, min_bucket) in self.min_liquidity_offset_history.buckets.iter().enumerate() {
@@ -2036,33 +2023,10 @@ mod bucketed_history {
 			// If the total valid points is smaller than 1.0 (i.e. 32 in our fixed-point scheme),
 			// treat it as if we were fully decayed.
 			const FULLY_DECAYED: u16 = BUCKET_FIXED_POINT_ONE * BUCKET_FIXED_POINT_ONE;
-			if total_valid_points_tracked.checked_shr(required_decays).unwrap_or(0) < FULLY_DECAYED.into() {
+			if total_valid_points_tracked < FULLY_DECAYED.into() {
 				return None;
 			}
 
-			Some((total_valid_points_tracked, required_decays))
-		}
-
-		#[inline]
-		pub(super) fn calculate_success_probability_times_billion<T: Time>(
-			&self, now: T, offset_history_last_updated: T, half_life: Duration,
-			params: &ProbabilisticScoringFeeParameters, amount_msat: u64, capacity_msat: u64
-		) -> Option<u64> {
-			// If historical penalties are enabled, we try to calculate a probability of success
-			// given our historical distribution of min- and max-liquidity bounds in a channel.
-			// To do so, we walk the set of historical liquidity bucket (min, max) combinations
-			// (where min_idx < max_idx, as having a minimum above our maximum is an invalid
-			// state). For each pair, we calculate the probability as if the bucket's corresponding
-			// min- and max- liquidity bounds were our current liquidity bounds and then multiply
-			// that probability by the weight of the selected buckets.
-			let payment_pos = amount_to_pos(amount_msat, capacity_msat);
-			if payment_pos >= POSITION_TICKS { return None; }
-
-			// Check if all our buckets are zero, once decayed and treat it as if we had no data. We
-			// don't actually use the decayed buckets, though, as that would lose precision.
-			let (total_valid_points_tracked, _)
-				= self.get_total_valid_points(now, offset_history_last_updated, half_life)?;
-
 			let mut cumulative_success_prob_times_billion = 0;
 			// Special-case the 0th min bucket - it generally means we failed a payment, so only
 			// consider the highest (i.e. largest-offset-from-max-capacity) max bucket for all
@@ -2918,19 +2882,9 @@ mod tests {
 		let usage = ChannelUsage { amount_msat: 896, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), u64::max_value());
 
-		// No decay
-		SinceEpoch::advance(Duration::from_secs(4));
-		let usage = ChannelUsage { amount_msat: 128, ..usage };
-		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 0);
-		let usage = ChannelUsage { amount_msat: 256, ..usage };
-		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 93);
-		let usage = ChannelUsage { amount_msat: 768, ..usage };
-		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 1_479);
-		let usage = ChannelUsage { amount_msat: 896, ..usage };
-		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), u64::max_value());
-
 		// Half decay (i.e., three-quarter life)
-		SinceEpoch::advance(Duration::from_secs(1));
+		SinceEpoch::advance(Duration::from_secs(5));
+		scorer.decay_liquidity_certainty(Duration::from_secs(5));
 		let usage = ChannelUsage { amount_msat: 128, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 22);
 		let usage = ChannelUsage { amount_msat: 256, ..usage };
@@ -2942,6 +2896,7 @@ mod tests {
 
 		// One decay (i.e., half life)
 		SinceEpoch::advance(Duration::from_secs(5));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10));
 		let usage = ChannelUsage { amount_msat: 64, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 0);
 		let usage = ChannelUsage { amount_msat: 128, ..usage };
@@ -2953,6 +2908,7 @@ mod tests {
 
 		// Fully decay liquidity lower bound.
 		SinceEpoch::advance(Duration::from_secs(10 * 7));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * 8));
 		let usage = ChannelUsage { amount_msat: 0, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 0);
 		let usage = ChannelUsage { amount_msat: 1, ..usage };
@@ -2964,12 +2920,14 @@ mod tests {
 
 		// Fully decay liquidity upper bound.
 		SinceEpoch::advance(Duration::from_secs(10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * 9));
 		let usage = ChannelUsage { amount_msat: 0, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 0);
 		let usage = ChannelUsage { amount_msat: 1_024, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), u64::max_value());
 
 		SinceEpoch::advance(Duration::from_secs(10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * 10));
 		let usage = ChannelUsage { amount_msat: 0, ..usage };
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 0);
 		let usage = ChannelUsage { amount_msat: 1_024, ..usage };
@@ -3004,9 +2962,11 @@ mod tests {
 		// An unchecked right shift 64 bits or more in DirectedChannelLiquidity::decayed_offset_msat
 		// would cause an overflow.
 		SinceEpoch::advance(Duration::from_secs(10 * 64));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * 64));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 125);
 
 		SinceEpoch::advance(Duration::from_secs(10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * 65));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 125);
 	}
 
@@ -3040,6 +3000,7 @@ mod tests {
 
 		// Decaying knowledge gives less confidence (128, 896), meaning a higher penalty.
 		SinceEpoch::advance(Duration::from_secs(10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 291);
 
 		// Reducing the upper bound gives more confidence (128, 832) that the payment amount (512)
@@ -3054,6 +3015,7 @@ mod tests {
 
 		// Further decaying affects the lower bound more than the upper bound (128, 928).
 		SinceEpoch::advance(Duration::from_secs(10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(20));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 280);
 	}
 
@@ -3083,6 +3045,7 @@ mod tests {
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), u64::max_value());
 
 		SinceEpoch::advance(Duration::from_secs(10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 473);
 
 		scorer.payment_path_failed(&payment_path_for_amount(250), 43, Duration::from_secs(10));
@@ -3097,8 +3060,7 @@ mod tests {
 		assert_eq!(deserialized_scorer.channel_penalty_msat(42, &source, &target, usage, &params), 300);
 	}
 
-	#[test]
-	fn decays_persisted_liquidity_bounds() {
+	fn do_decays_persisted_liquidity_bounds(decay_before_reload: bool) {
 		let logger = TestLogger::new();
 		let network_graph = network_graph(&logger);
 		let params = ProbabilisticScoringFeeParameters {
@@ -3122,23 +3084,38 @@ mod tests {
 		scorer.payment_path_failed(&payment_path_for_amount(500), 42, Duration::ZERO);
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), u64::max_value());
 
+		if decay_before_reload {
+			SinceEpoch::advance(Duration::from_secs(10));
+			scorer.decay_liquidity_certainty(Duration::from_secs(10));
+		}
+
 		let mut serialized_scorer = Vec::new();
 		scorer.write(&mut serialized_scorer).unwrap();
 
-		SinceEpoch::advance(Duration::from_secs(10));
-
 		let mut serialized_scorer = io::Cursor::new(&serialized_scorer);
-		let deserialized_scorer =
+		let mut deserialized_scorer =
 			<ProbabilisticScorer>::read(&mut serialized_scorer, (decay_params, &network_graph, &logger)).unwrap();
+		if !decay_before_reload {
+			SinceEpoch::advance(Duration::from_secs(10));
+			scorer.decay_liquidity_certainty(Duration::from_secs(10));
+			deserialized_scorer.decay_liquidity_certainty(Duration::from_secs(10));
+		}
 		assert_eq!(deserialized_scorer.channel_penalty_msat(42, &source, &target, usage, &params), 473);
 
 		scorer.payment_path_failed(&payment_path_for_amount(250), 43, Duration::from_secs(10));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 300);
 
 		SinceEpoch::advance(Duration::from_secs(10));
+		deserialized_scorer.decay_liquidity_certainty(Duration::from_secs(20));
 		assert_eq!(deserialized_scorer.channel_penalty_msat(42, &source, &target, usage, &params), 370);
 	}
 
+	#[test]
+	fn decays_persisted_liquidity_bounds() {
+		do_decays_persisted_liquidity_bounds(false);
+		do_decays_persisted_liquidity_bounds(true);
+	}
+
 	#[test]
 	fn scores_realistic_payments() {
 		// Shows the scores of "realistic" sends of 100k sats over channels of 1-10m sats (with a
@@ -3402,11 +3379,11 @@ mod tests {
 		// Advance the time forward 16 half-lives (which the docs claim will ensure all data is
 		// gone), and check that we're back to where we started.
 		SinceEpoch::advance(Duration::from_secs(10 * 16));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * 16));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 168);
-		// Once fully decayed we still have data, but its all-0s. In the future we may remove the
-		// data entirely instead.
+		// Once fully decayed we still have data, so we get 0s for the buckets.
 		assert_eq!(scorer.historical_estimated_channel_liquidity_probabilities(42, &target),
-			None);
+			Some(([0; 32], [0; 32])));
 		assert_eq!(scorer.historical_estimated_payment_success_probability(42, &target, 1, &params), None);
 
 		let mut usage = ChannelUsage {
@@ -3417,7 +3394,7 @@ mod tests {
 		scorer.payment_path_failed(&payment_path_for_amount(1), 42, Duration::from_secs(10 * 16));
 		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 2050);
 		usage.inflight_htlc_msat = 0;
-		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 866);
+		assert_eq!(scorer.channel_penalty_msat(42, &source, &target, usage, &params), 2048);
 
 		let usage = ChannelUsage {
 			amount_msat: 1,
@@ -3428,6 +3405,12 @@ mod tests {
 
 		// Advance to decay all liquidity offsets to zero.
 		SinceEpoch::advance(Duration::from_secs(60 * 60 * 10));
+		scorer.decay_liquidity_certainty(Duration::from_secs(10 * (16 + 60 * 60)));
+
+		// Once even the bounds have decayed information about the channel should be removed
+		// entirely.
+		assert_eq!(scorer.historical_estimated_channel_liquidity_probabilities(42, &target),
+			None);
 
 		// Use a path in the opposite direction, which have zero for htlc_maximum_msat. This will
 		// ensure that the effective capacity is zero to test division-by-zero edge cases.