Support cross-epoch leader reputation (aptos-labs#2603)

Description Instead of restarting history on every epoch, we stop special treating epoch boundary, and look at the window size history. In order to do so, we need to know historical validator sets, to understand indices in the arrays, and that is the main complexity in this diff Also removing ActiveInactive leader reputation heuristic, as it is unnecessary and replaced by the ProposerAndVoter heuristic. (we kept the old one for the rollout of ProposerAndVoter, in case we need to fall back, time to clean it up) Test Plan leader_reputation and aptos_cli.test_join_and_leave_validator tests
matt-am · Aug 10, 2022 · 515e072 · 515e072
1 parent 72d8186
commit 515e072
Show file tree

Hide file tree

Showing 8 changed files with 437 additions and 351 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/...ts_test__test_get_account_resources_returns_empty_array_for_account_has_no_resources.json b/...ts_test__test_get_account_resources_returns_empty_array_for_account_has_no_resources.json
@@ -865,7 +865,7 @@
   {
     "type": "0x1::consensus_config::ConsensusConfig",
     "data": {
-      "config": "0x00010a0000000000000014000000000000000201e8030000000000000a0000000000000001000000000000000a0000000a000000000000000100000000000000010a00000000000000"
+      "config": "0x00010a0000000000000014000000000000000200e8030000000000000a0000000000000001000000000000000a0000000a00000000000000010000000000000001050000000a00000000000000"
     }
   }
 ]
diff --git a/...ts_test__test_get_account_resources_returns_empty_array_for_account_has_no_resources.json b/...ts_test__test_get_account_resources_returns_empty_array_for_account_has_no_resources.json
@@ -865,7 +865,7 @@
   {
     "type": "0x1::consensus_config::ConsensusConfig",
     "data": {
-      "config": "0x00010a0000000000000014000000000000000201e8030000000000000a0000000000000001000000000000000a0000000a000000000000000100000000000000010a00000000000000"
+      "config": "0x00010a0000000000000014000000000000000200e8030000000000000a0000000000000001000000000000000a0000000a00000000000000010000000000000001050000000a00000000000000"
     }
   }
 ]
diff --git a/consensus/Cargo.toml b/consensus/Cargo.toml
@@ -32,6 +32,7 @@ tokio = { version = "1.18.2", features = ["full"] }
 aptos-config = { path = "../config" }
 aptos-crypto = { path = "../crates/aptos-crypto" }
 aptos-infallible = { path = "../crates/aptos-infallible" }
+aptos-keygen = { path = "../crates/aptos-keygen" }
 aptos-logger = { path = "../crates/aptos-logger" }
 aptos-mempool = { path = "../mempool" }
 aptos-metrics-core = { path = "../crates/aptos-metrics-core" }

diff --git a/consensus/src/epoch_manager.rs b/consensus/src/epoch_manager.rs
@@ -1,6 +1,7 @@
 // Copyright (c) Aptos
 // SPDX-License-Identifier: Apache-2.0
 
+use crate::liveness::leader_reputation::extract_epoch_to_proposers;
 use crate::monitor;
 use crate::{
     block_storage::BlockStore,
@@ -15,8 +16,7 @@ use crate::{
     liveness::{
         cached_proposer_election::CachedProposerElection,
         leader_reputation::{
-            ActiveInactiveHeuristic, AptosDBBackend, LeaderReputation, ProposerAndVoterHeuristic,
-            ReputationHeuristic,
+            AptosDBBackend, LeaderReputation, ProposerAndVoterHeuristic, ReputationHeuristic,
         },
         proposal_generator::ProposalGenerator,
         proposer_election::ProposerElection,
@@ -65,8 +65,10 @@ use futures::{
     },
     SinkExt, StreamExt,
 };
+use itertools::Itertools;
 use network::protocols::network::{ApplicationNetworkSender, Event};
 use safety_rules::SafetyRulesManager;
+use std::collections::HashMap;
 use std::{
     cmp::Ordering,
     mem::{discriminant, Discriminant},
@@ -205,24 +207,12 @@ impl EpochManager {
                 Box::new(RotatingProposer::new(vec![proposer], *contiguous_rounds))
             }
             ProposerElectionType::LeaderReputation(leader_reputation_type) => {
-                let (heuristic, window_size, weight_by_voting_power) = match &leader_reputation_type
-                {
-                    LeaderReputationType::ActiveInactive(active_inactive_config) => {
-                        let window_size = proposers.len()
-                            * active_inactive_config.window_num_validators_multiplier;
-                        let heuristic: Box<dyn ReputationHeuristic> =
-                            Box::new(ActiveInactiveHeuristic::new(
-                                self.author,
-                                active_inactive_config.active_weight,
-                                active_inactive_config.inactive_weight,
-                                window_size,
-                            ));
-                        (
-                            heuristic,
-                            window_size,
-                            active_inactive_config.weight_by_voting_power,
-                        )
-                    }
+                let (
+                    heuristic,
+                    window_size,
+                    weight_by_voting_power,
+                    use_history_from_previous_epoch_max_count,
+                ) = match &leader_reputation_type {
                     LeaderReputationType::ProposerAndVoter(proposer_and_voter_config) => {
                         let proposer_window_size = proposers.len()
                             * proposer_and_voter_config.proposer_window_num_validators_multiplier;
@@ -242,16 +232,18 @@ impl EpochManager {
                             heuristic,
                             std::cmp::max(proposer_window_size, voter_window_size),
                             proposer_and_voter_config.weight_by_voting_power,
+                            proposer_and_voter_config.use_history_from_previous_epoch_max_count,
                         )
                     }
                 };
 
+                let seek_len = onchain_config.leader_reputation_exclude_round() as usize
+                    + onchain_config.max_failed_authors_to_store()
+                    + PROPSER_ROUND_BEHIND_STORAGE_BUFFER;
+
                 let backend = Box::new(AptosDBBackend::new(
-                    epoch_state.epoch,
                     window_size,
-                    onchain_config.leader_reputation_exclude_round() as usize
-                        + onchain_config.max_failed_authors_to_store()
-                        + PROPSER_ROUND_BEHIND_STORAGE_BUFFER,
+                    seek_len,
                     self.storage.aptos_db(),
                 ));
                 let voting_powers: Vec<_> = if weight_by_voting_power {
@@ -263,9 +255,43 @@ impl EpochManager {
                     vec![1; proposers.len()]
                 };
 
+                // First block (after genesis) is epoch=1, so that is the first epoch we consider. (Genesis is epoch=0)
+                let first_epoch_to_consider = std::cmp::max(
+                    1,
+                    epoch_state
+                        .epoch
+                        .saturating_sub(use_history_from_previous_epoch_max_count as u64),
+                );
+                // If we are considering beyond the current epoch, we need to fetch validators for those epochs
+                let epoch_to_proposers = if epoch_state.epoch > first_epoch_to_consider {
+                    self.storage
+                        .aptos_db()
+                        .get_epoch_ending_ledger_infos(first_epoch_to_consider - 1, epoch_state.epoch)
+                        .and_then(|proof| {
+                            ensure!(proof.ledger_info_with_sigs.len() as u64 == (epoch_state.epoch - (first_epoch_to_consider - 1)));
+                            extract_epoch_to_proposers(proof, epoch_state.epoch, &proposers, (window_size + seek_len) as u64)
+                        })
+                        .unwrap_or_else(|err| {
+                            error!("Couldn't create leader reputation with history across epochs, {:?}", err);
+                            HashMap::from([(epoch_state.epoch, proposers)])
+                        })
+                } else {
+                    HashMap::from([(epoch_state.epoch, proposers)])
+                };
+
+                info!(
+                    "Starting epoch {}: proposers across epochs for leader election: {:?}",
+                    epoch_state.epoch,
+                    epoch_to_proposers
+                        .iter()
+                        .map(|(epoch, proposers)| (epoch, proposers.len()))
+                        .sorted()
+                        .collect::<Vec<_>>()
+                );
+
                 let proposer_election = Box::new(LeaderReputation::new(
                     epoch_state.epoch,
-                    proposers,
+                    epoch_to_proposers,
                     voting_powers,
                     backend,
                     heuristic,