bugfix: manually calculate per node work on rewarded set changes (#5972)

common/cosmwasm-smart-contracts/mixnet-contract/src/reward_params.rs

@@ -86,6 +86,25 @@ impl IntervalRewardParams {
     pub fn to_inline_json(&self) -> String {
         to_json_string(self).unwrap_or_else(|_| "serialisation failure".into())
     }
+
+    pub fn active_node_work(&self, standby_node_work: Decimal) -> WorkFactor {
+        self.active_set_work_factor * standby_node_work
+    }
+
+    pub fn standby_node_work(
+        &self,
+        rewarded_set_size: Decimal,
+        standby_set_size: Decimal,
+    ) -> WorkFactor {
+        let f = self.active_set_work_factor;
+        let k = rewarded_set_size;
+        let one = Decimal::one();
+
+        // nodes in reserve
+        let k_r = standby_set_size;
+
+        one / (f * k - (f - one) * k_r)
+    }
 }
 
 /// Parameters used for reward calculation.
@@ -109,18 +128,15 @@ pub struct RewardingParams {
 
 impl RewardingParams {
     pub fn active_node_work(&self) -> WorkFactor {
-        self.interval.active_set_work_factor * self.standby_node_work()
+        let standby_work = self.standby_node_work();
+        self.interval.active_node_work(standby_work)
     }
 
     pub fn standby_node_work(&self) -> WorkFactor {
-        let f = self.interval.active_set_work_factor;
-        let k = self.dec_rewarded_set_size();
-        let one = Decimal::one();
-
-        // nodes in reserve
-        let k_r = self.dec_standby_set_size();
-
-        one / (f * k - (f - one) * k_r)
+        let rewarded_set_size = self.dec_rewarded_set_size();
+        let standby_set_size = self.dec_standby_set_size();
+        self.interval
+            .standby_node_work(rewarded_set_size, standby_set_size)
     }
 
     pub fn rewarded_set_size(&self) -> u32 {

common/cosmwasm-smart-contracts/mixnet-contract/src/types.rs

@@ -3,6 +3,7 @@
 
 use crate::config_score::{ConfigScoreParams, OutdatedVersionWeights, VersionScoreFormulaParams};
 use crate::nym_node::Role;
+use crate::reward_params::RewardedSetParams;
 use crate::EpochId;
 use contracts_common::Percent;
 use cosmwasm_schema::cw_serde;
@@ -85,7 +86,11 @@ impl RewardedSet {
     }
 
     pub fn rewarded_set_size(&self) -> usize {
-        self.active_set_size() + self.standby.len()
+        self.active_set_size() + self.standby_set_size()
+    }
+
+    pub fn standby_set_size(&self) -> usize {
+        self.standby.len()
     }
 
     pub fn get_role(&self, node_id: NodeId) -> Option<Role> {
@@ -110,6 +115,13 @@ impl RewardedSet {
         }
         None
     }
+
+    pub fn matches_parameters(&self, params: RewardedSetParams) -> bool {
+        self.entry_gateways.len() <= params.entry_gateways as usize
+            && self.exit_gateways.len() <= params.exit_gateways as usize
+            && self.layer1.len() + self.layer2.len() + self.layer3.len() <= params.mixnodes as usize
+            && self.standby.len() <= params.standby as usize
+    }
 }
 
 #[cw_serde]

nym-api/src/epoch_operations/helpers.rs

@@ -5,12 +5,16 @@ use crate::epoch_operations::EpochAdvancer;
 use crate::support::caching::Cache;
 use cosmwasm_std::{Decimal, Fraction};
 use nym_api_requests::models::NodeAnnotation;
+use nym_mixnet_contract_common::helpers::IntoBaseDecimal;
 use nym_mixnet_contract_common::reward_params::{NodeRewardingParameters, Performance, WorkFactor};
-use nym_mixnet_contract_common::{EpochRewardedSet, ExecuteMsg, NodeId, RewardingParams};
+use nym_mixnet_contract_common::{
+    EpochRewardedSet, ExecuteMsg, NodeId, RewardedSet, RewardingParams,
+};
 use serde::{Deserialize, Serialize};
+use std::cmp::max;
 use std::collections::HashMap;
 use tokio::sync::RwLockReadGuard;
-use tracing::error;
+use tracing::{debug, error};
 
 #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
 pub(crate) struct NodeWithPerformance {
@@ -73,6 +77,137 @@ pub(super) fn stake_to_f64(stake: Decimal) -> f64 {
     }
 }
 
+struct PerNodeWork {
+    active: WorkFactor,
+    standby: WorkFactor,
+}
+
+struct NodeWorkCalculationComponents {
+    active_set_size: Decimal,
+    standby_set_size: Decimal,
+    per_node_work: PerNodeWork,
+}
+
+impl NodeWorkCalculationComponents {
+    fn standby_set_work_share(&self) -> Decimal {
+        self.standby_set_size * self.per_node_work.standby
+    }
+
+    fn active_set_work_share(&self) -> Decimal {
+        self.active_set_size * self.per_node_work.active
+    }
+}
+
+fn default_node_work_calculation(
+    nodes: &RewardedSet,
+    global_rewarding_params: RewardingParams,
+) -> NodeWorkCalculationComponents {
+    let per_node_work = PerNodeWork {
+        active: global_rewarding_params.active_node_work(),
+        standby: global_rewarding_params.standby_node_work(),
+    };
+    // SANITY CHECK:
+    // SAFETY: 0 decimal places is within the range of `Decimal`
+    #[allow(clippy::unwrap_used)]
+    let standby_set_size = Decimal::from_atomics(nodes.standby.len() as u128, 0).unwrap();
+    #[allow(clippy::unwrap_used)]
+    let active_set_size = Decimal::from_atomics(nodes.active_set_size() as u128, 0).unwrap();
+
+    NodeWorkCalculationComponents {
+        active_set_size,
+        standby_set_size,
+        per_node_work,
+    }
+}
+
+fn manual_node_work_calculation(
+    nodes: &RewardedSet,
+    global_rewarding_params: RewardingParams,
+) -> NodeWorkCalculationComponents {
+    // calculate everything manually based on the actual rewarded set on hand
+    // but always attempt to minimise the node work, so take the maximum values
+    // of the set sizes between new and old parameters
+    // (more nodes = smaller per-node work as it has to be spread through more entries)
+    let rewarded_set_size = max(
+        global_rewarding_params.rewarded_set.rewarded_set_size(),
+        nodes.rewarded_set_size() as u32,
+    );
+    let standby_set_size = max(
+        global_rewarding_params.rewarded_set.standby,
+        nodes.standby_set_size() as u32,
+    );
+    // the unwraps here are fine as we're guaranteed an `u32` is going to fit in a Decimal with 0 decimal places
+    #[allow(clippy::unwrap_used)]
+    let rewarded_set_size_dec = rewarded_set_size.into_base_decimal().unwrap();
+    #[allow(clippy::unwrap_used)]
+    let standby_set_size_dec = standby_set_size.into_base_decimal().unwrap();
+    #[allow(clippy::unwrap_used)]
+    let active_set_size = rewarded_set_size
+        .saturating_sub(standby_set_size)
+        .into_base_decimal()
+        .unwrap();
+
+    let standby_node_work = global_rewarding_params
+        .interval
+        .standby_node_work(rewarded_set_size_dec, standby_set_size_dec);
+    let active_node_work = global_rewarding_params
+        .interval
+        .active_node_work(standby_node_work);
+    let per_node_work = PerNodeWork {
+        active: active_node_work,
+        standby: standby_node_work,
+    };
+
+    NodeWorkCalculationComponents {
+        active_set_size,
+        standby_set_size: standby_set_size_dec,
+        per_node_work,
+    }
+}
+
+fn determine_per_node_work(
+    nodes: &RewardedSet,
+    // we only need reward parameters for active set work factor and rewarded/active set sizes;
+    // we do not need exact values of reward pool, staking supply, etc., so it's fine if it's slightly out of sync
+    global_rewarding_params: RewardingParams,
+) -> PerNodeWork {
+    // currently we are using constant omega for nodes, but that will change with tickets
+    // or different reward split between entry, exit, etc. at that point this will have to be calculated elsewhere
+    let res = if nodes.matches_parameters(global_rewarding_params.rewarded_set) {
+        default_node_work_calculation(nodes, global_rewarding_params)
+    } else {
+        error!("the current rewarded set does not much current rewarding parameters. this could only be expected if rewarded set distribution has been changed mid-epoch");
+        manual_node_work_calculation(nodes, global_rewarding_params)
+    };
+
+    let active_node_work_factor = res.per_node_work.active;
+    let standby_node_work_factor = res.per_node_work.standby;
+
+    debug!("using {active_node_work_factor} as active node work factor and {standby_node_work_factor} as standby node work factor");
+
+    let standby_share = res.standby_set_work_share();
+    let active_share = res.active_set_work_share();
+    let total_work = standby_share + active_share;
+
+    // this HAS TO blow up. there's no recovery
+    #[allow(clippy::panic)]
+    if total_work > Decimal::one() {
+        panic!("work calculation logic is flawed! somehow the total work in the system is greater than 1! \
+            total work={total_work}, \
+            active set share={active_share}, \
+            standby share={standby_share}, \
+            active node work factor={active_node_work_factor}, \
+            standby node work factor={standby_node_work_factor}, \
+            active set size={} \
+            standby set size={}", res.active_set_size, res.standby_set_size);
+    }
+
+    PerNodeWork {
+        active: active_node_work_factor,
+        standby: standby_node_work_factor,
+    }
+}
+
 impl EpochAdvancer {
     fn load_performance(
         status_cache: &Option<RwLockReadGuard<Cache<HashMap<NodeId, NodeAnnotation>>>>,
@@ -99,23 +234,9 @@ impl EpochAdvancer {
         global_rewarding_params: RewardingParams,
     ) -> Vec<RewardedNodeWithParams> {
         let nodes = &nodes.assignment;
-        // currently we are using constant omega for nodes, but that will change with tickets
-        // or different reward split between entry, exit, etc. at that point this will have to be calculated elsewhere
-        let active_node_work_factor = global_rewarding_params.active_node_work();
-        let standby_node_work_factor = global_rewarding_params.standby_node_work();
-
-        // SANITY CHECK:
-        // SAFETY: 0 decimal places is within the range of `Decimal`
-        #[allow(clippy::unwrap_used)]
-        let standby_share = Decimal::from_atomics(nodes.standby.len() as u128, 0).unwrap()
-            * standby_node_work_factor;
-        #[allow(clippy::unwrap_used)]
-        let active_share = Decimal::from_atomics(nodes.active_set_size() as u128, 0).unwrap()
-            * active_node_work_factor;
-        let total_work = standby_share + active_share;
-
-        // this HAS TO blow up. there's no recovery
-        assert!(total_work <= Decimal::one(), "work calculation logic is flawed! somehow the total work in the system is greater than 1!");
+        let nodes_work = determine_per_node_work(nodes, global_rewarding_params);
+        let active_node_work_factor = nodes_work.active;
+        let standby_node_work_factor = nodes_work.standby;
 
         let status_cache = self.status_cache.node_annotations().await;
         if status_cache.is_none() {
@@ -161,6 +282,9 @@ impl EpochAdvancer {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use nym_contracts_common::Percent;
+    use nym_mixnet_contract_common::reward_params::RewardedSetParams;
+    use nym_mixnet_contract_common::IntervalRewardParams;
 
     fn compare_large_floats(a: f64, b: f64) {
         // for very large floats, allow for smaller larger epsilon
@@ -206,4 +330,72 @@ mod tests {
             compare_large_floats(expected_f64, stake_to_f64(decimal))
         }
     }
+
+    fn dummy_rewarding_params() -> RewardingParams {
+        RewardingParams {
+            interval: IntervalRewardParams {
+                reward_pool: Decimal::from_atomics(100_000_000_000_000u128, 0).unwrap(),
+                staking_supply: Decimal::from_atomics(123_456_000_000_000u128, 0).unwrap(),
+                staking_supply_scale_factor: Percent::hundred(),
+                epoch_reward_budget: Decimal::from_ratio(100_000_000_000_000u128, 1234u32)
+                    * Decimal::percent(1),
+                stake_saturation_point: Decimal::from_ratio(123_456_000_000_000u128, 313u32),
+                sybil_resistance: Percent::from_percentage_value(23).unwrap(),
+                active_set_work_factor: Decimal::from_atomics(10u32, 0).unwrap(),
+                interval_pool_emission: Percent::from_percentage_value(1).unwrap(),
+            },
+            rewarded_set: RewardedSetParams {
+                entry_gateways: 50,
+                exit_gateways: 70,
+                mixnodes: 120,
+                standby: 20,
+            },
+        }
+    }
+
+    #[test]
+    fn determining_nodes_work() {
+        let params = dummy_rewarding_params();
+        // matched parameters
+        let rewarded_set = RewardedSet {
+            entry_gateways: (1..)
+                .take(params.rewarded_set.entry_gateways as usize)
+                .collect(),
+            exit_gateways: (1000..)
+                .take(params.rewarded_set.exit_gateways as usize)
+                .collect(),
+            layer1: (2000..)
+                .take(params.rewarded_set.mixnodes as usize / 3)
+                .collect(),
+            layer2: (3000..)
+                .take(params.rewarded_set.mixnodes as usize / 3)
+                .collect(),
+            layer3: (4000..)
+                .take(params.rewarded_set.mixnodes as usize / 3)
+                .collect(),
+            standby: (5000..)
+                .take(params.rewarded_set.standby as usize)
+                .collect(),
+        };
+
+        let work = determine_per_node_work(&rewarded_set, params);
+        assert_eq!(work.active, params.active_node_work());
+        assert_eq!(work.standby, params.standby_node_work());
+
+        // updated
+        // here we're interested in the fact that the calculation does not panic, i.e. total work <= 1
+        let params = dummy_rewarding_params();
+        let rewarded_set = RewardedSet {
+            entry_gateways: (1..).take(250).collect(),
+            exit_gateways: (1000..).take(100).collect(),
+            layer1: (2000..).take(10).collect(),
+            layer2: (3000..).take(10).collect(),
+            layer3: (4000..).take(10).collect(),
+            standby: (5000..).take(5).collect(),
+        };
+
+        let work = determine_per_node_work(&rewarded_set, params);
+        assert_ne!(work.active, params.active_node_work());
+        assert_ne!(work.standby, params.standby_node_work());
+    }
 }