wip

Author: akhi3030

Cargo.lock

@@ -70,6 +70,7 @@ lru = { workspace = true }
 min-max-heap = { workspace = true }
 num_cpus = { workspace = true }
 num_enum = { workspace = true }
+parking_lot = { workspace = true }
 prio-graph = { workspace = true }
 qualifier_attr = { workspace = true }
 quinn = { workspace = true }

core/Cargo.toml

@@ -9,6 +9,7 @@ use {
         banking_trace::BankingTracer,
         replay_stage::{Finalizer, ReplayStage},
     },
+    parking_lot::RwLock as PLRwLock,
     solana_clock::Slot,
     solana_entry::block_component::{
         BlockFooterV1, BlockMarkerV1, VersionedBlockFooter, VersionedBlockMarker,
@@ -31,8 +32,14 @@ use {
         bank_forks::BankForks,
     },
     solana_version::version,
-    solana_votor::{common::block_timeout, event::LeaderWindowInfo, votor::LeaderWindowNotifier},
-    solana_votor_messages::migration::MigrationStatus,
+    solana_votor::{
+        common::block_timeout, consensus_rewards::ConsensusRewards, event::LeaderWindowInfo,
+        votor::LeaderWindowNotifier,
+    },
+    solana_votor_messages::{
+        migration::MigrationStatus,
+        rewards_certificate::{NotarRewardCertificate, SkipRewardCertificate},
+    },
     stats::{BlockCreationLoopMetrics, SlotMetrics},
     std::{
         sync::{
@@ -79,6 +86,7 @@ pub struct BlockCreationLoopConfig {
     pub poh_recorder: Arc<RwLock<PohRecorder>>,
     pub leader_schedule_cache: Arc<LeaderScheduleCache>,
     pub rpc_subscriptions: Option<Arc<RpcSubscriptions>>,
+    pub consensus_rewards: Arc<PLRwLock<ConsensusRewards>>,
 
     // Notifiers
     pub banking_tracer: Arc<BankingTracer>,
@@ -104,6 +112,7 @@ struct LeaderContext {
     slot_status_notifier: Option<SlotStatusNotifier>,
     banking_tracer: Arc<BankingTracer>,
     replay_highest_frozen: Arc<ReplayHighestFrozen>,
+    consensus_rewards: Arc<PLRwLock<ConsensusRewards>>,
 
     // Metrics
     metrics: BlockCreationLoopMetrics,
@@ -134,9 +143,15 @@ enum StartLeaderError {
     ),
 }
 
-fn produce_block_footer(block_producer_time_nanos: u64) -> VersionedBlockMarker {
+fn produce_block_footer(
+    block_producer_time_nanos: u64,
+    skip_reward_certificate: Option<SkipRewardCertificate>,
+    notar_reward_certificate: Option<NotarRewardCertificate>,
+) -> VersionedBlockMarker {
     let footer = BlockFooterV1 {
         block_producer_time_nanos,
+        skip_reward_certificate,
+        notar_reward_certificate,
         block_user_agent: format!("agave/{}", version!()).into_bytes(),
     };
 
@@ -166,6 +181,7 @@ fn start_loop(config: BlockCreationLoopConfig) {
         leader_window_notifier,
         replay_highest_frozen,
         migration_status,
+        consensus_rewards,
     } = config;
 
     // Similar to Votor, if this loop dies kill the validator
@@ -181,14 +197,15 @@ fn start_loop(config: BlockCreationLoopConfig) {
         blockstore,
         poh_recorder: poh_recorder.clone(),
         record_receiver,
-        leader_schedule_cache,
+        leader_schedule_cache: leader_schedule_cache.clone(),
         bank_forks,
         rpc_subscriptions,
         slot_status_notifier,
         banking_tracer,
         replay_highest_frozen,
         metrics: BlockCreationLoopMetrics::default(),
         slot_metrics: SlotMetrics::default(),
+        consensus_rewards,
     };
 
     info!("{my_pubkey}: Block creation loop initialized");
@@ -301,6 +318,8 @@ fn produce_window(
         if let Err(e) = record_and_complete_block(
             ctx.poh_recorder.as_ref(),
             &mut ctx.record_receiver,
+            // XXX: currently this can content with the write lock in sigverifier.  Fix that.
+            &ctx.consensus_rewards.read(),
             skip_timer,
             timeout,
         ) {
@@ -342,6 +361,7 @@ fn produce_window(
 fn record_and_complete_block(
     poh_recorder: &RwLock<PohRecorder>,
     record_receiver: &mut RecordReceiver,
+    consensus_rewards: &ConsensusRewards,
     block_timer: Instant,
     block_timeout: Duration,
 ) -> Result<(), PohRecorderError> {
@@ -378,7 +398,10 @@ fn record_and_complete_block(
     // Construct and send the block footer
     let mut w_poh_recorder = poh_recorder.write().unwrap();
     let block_producer_time_nanos = w_poh_recorder.working_bank_block_producer_time_nanos();
-    let footer = produce_block_footer(block_producer_time_nanos);
+    // XXX: how to look up the slot.
+    let slot = u64::MAX;
+    let (skip, notar) = consensus_rewards.build_rewards_certs(slot);
+    let footer = produce_block_footer(block_producer_time_nanos, skip, notar);
     w_poh_recorder.send_marker(footer)?;
 
     // Alpentick and clear bank

core/src/block_creation_loop.rs

@@ -9,6 +9,7 @@ use {
     },
     bitvec::prelude::{BitVec, Lsb0},
     crossbeam_channel::{Sender, TrySendError},
+    parking_lot::RwLock as PLRwLock,
     rayon::iter::{
         IndexedParallelIterator, IntoParallelIterator, IntoParallelRefIterator, ParallelIterator,
     },
@@ -23,7 +24,10 @@ use {
     solana_runtime::{bank::Bank, bank_forks::SharableBanks, epoch_stakes::BLSPubkeyToRankMap},
     solana_signer_store::{decode, DecodeError},
     solana_streamer::packet::PacketBatch,
-    solana_votor::consensus_metrics::{ConsensusMetricsEvent, ConsensusMetricsEventSender},
+    solana_votor::{
+        consensus_metrics::{ConsensusMetricsEvent, ConsensusMetricsEventSender},
+        consensus_rewards::ConsensusRewards,
+    },
     solana_votor_messages::{
         consensus_message::{Certificate, CertificateType, ConsensusMessage, VoteMessage},
         vote::Vote,
@@ -92,6 +96,7 @@ pub struct BLSSigVerifier {
     consensus_metrics_sender: ConsensusMetricsEventSender,
     last_checked_root_slot: Slot,
     alpenglow_last_voted: Arc<AlpenglowLastVoted>,
+    consensus_rewards: Arc<PLRwLock<ConsensusRewards>>,
 }
 
 impl BLSSigVerifier {
@@ -172,8 +177,17 @@ impl BLSSigVerifier {
                         id: *solana_pubkey,
                         vote: vote_message.vote,
                     });
-                    // Only need votes newer than root slot
+
+                    // consensus pool does not need votes for slots other than root slot however the rewards container may still need them.
                     if vote_message.vote.slot() <= root_bank.slot() {
+                        // the only module that takes a write lock on consensus_rewards is this one and it does not take the write lock while it is verifying votes so this should never block.
+                        if self
+                            .consensus_rewards
+                            .read()
+                            .wants_vote(root_bank.slot(), &vote_message)
+                        {
+                            // XXX: actually verify and send the votes.  The verification and sending should happen off the critical path.
+                        }
                         self.stats.received_old.fetch_add(1, Ordering::Relaxed);
                         packet.meta_mut().set_discard(true);
                         continue;
@@ -219,8 +233,8 @@ impl BLSSigVerifier {
             || self.verify_and_send_certificates(certs_to_verify, &root_bank),
         );
 
-        votes_result?;
-        certs_result?;
+        let rewards_votes = votes_result?;
+        let () = certs_result?;
 
         // Send to RPC service for last voted tracking
         self.alpenglow_last_voted
@@ -235,6 +249,16 @@ impl BLSSigVerifier {
             warn!("could not send consensus metrics, receive side of channel is closed");
         }
 
+        {
+            // Grabbing this lock should not contend with this module as this module takes the read lock during the verification phase which should be over by now.
+            // XXX: this can contend with the block creation loop.  Consider using try_write() instead or other approaches to minimise it.
+            let mut guard = self.consensus_rewards.write();
+            let root_slot = root_bank.slot();
+            for v in rewards_votes {
+                guard.add_vote_message(root_slot, v);
+            }
+        }
+
         self.stats.maybe_report_stats();
 
         Ok(())
@@ -248,6 +272,7 @@ impl BLSSigVerifier {
         message_sender: Sender<ConsensusMessage>,
         consensus_metrics_sender: ConsensusMetricsEventSender,
         alpenglow_last_voted: Arc<AlpenglowLastVoted>,
+        consensus_rewards: Arc<PLRwLock<ConsensusRewards>>,
     ) -> Self {
         Self {
             sharable_banks,
@@ -259,14 +284,32 @@ impl BLSSigVerifier {
             consensus_metrics_sender,
             last_checked_root_slot: 0,
             alpenglow_last_voted,
+            consensus_rewards,
         }
     }
 
+    /// Verifies votes and sends verified votes to the consensus pool.
+    /// Also returns a copy of the verified votes that the rewards container is intersted is so that the caller can send them to it.
     fn verify_and_send_votes(
         &self,
         votes_to_verify: Vec<VoteToVerify>,
-    ) -> Result<(), BLSSigVerifyServiceError<ConsensusMessage>> {
+    ) -> Result<Vec<VoteMessage>, BLSSigVerifyServiceError<ConsensusMessage>> {
         let verified_votes = self.verify_votes(votes_to_verify);
+
+        let rewards_votes = {
+            // the only module that takes a write lock on consensus_rewards is this one and it does not take the write lock while it is verifying votes so this should never block.
+            let guard = self.consensus_rewards.read();
+            let root_slot = self.sharable_banks.root().slot();
+            verified_votes
+                .iter()
+                .filter_map(|vote| {
+                    guard
+                        .wants_vote(root_slot, &vote.vote_message)
+                        .then_some(vote.vote_message.clone())
+                })
+                .collect()
+        };
+
         self.stats
             .total_valid_packets
             .fetch_add(verified_votes.len() as u64, Ordering::Relaxed);
@@ -285,7 +328,7 @@ impl BLSSigVerifier {
                 }
             }
 
-            // Send the BLS vote messaage to certificate pool
+            // Send the votes to the consensus pool
             match self
                 .message_sender
                 .try_send(ConsensusMessage::Vote(vote.vote_message))
@@ -319,7 +362,7 @@ impl BLSSigVerifier {
             }
         }
 
-        Ok(())
+        Ok(rewards_votes)
     }
 
     fn verify_votes(&self, votes_to_verify: Vec<VoteToVerify>) -> Vec<VoteToVerify> {

core/src/bls_sigverify/bls_sigverifier.rs

@@ -27,6 +27,7 @@ use {
     },
     bytes::Bytes,
     crossbeam_channel::{bounded, unbounded, Receiver, Sender},
+    parking_lot::RwLock as PLRwLock,
     solana_client::connection_cache::ConnectionCache,
     solana_clock::Slot,
     solana_geyser_plugin_manager::block_metadata_notifier_interface::BlockMetadataNotifierArc,
@@ -60,6 +61,7 @@ use {
     },
     solana_turbine::{retransmit_stage::RetransmitStage, xdp::XdpSender},
     solana_votor::{
+        consensus_rewards::ConsensusRewards,
         event::{VotorEventReceiver, VotorEventSender},
         vote_history::VoteHistory,
         vote_history_storage::VoteHistoryStorage,
@@ -214,6 +216,7 @@ impl Tvu {
         key_notifiers: Arc<RwLock<KeyUpdaters>>,
         alpenglow_last_voted: Arc<AlpenglowLastVoted>,
         migration_status: Arc<MigrationStatus>,
+        consensus_rewards: Arc<PLRwLock<ConsensusRewards>>,
     ) -> Result<Self, String> {
         let (consensus_message_sender, consensus_message_receiver) =
             bounded(MAX_ALPENGLOW_PACKET_NUM);
@@ -274,6 +277,7 @@ impl Tvu {
                 consensus_message_sender.clone(),
                 consensus_metrics_sender.clone(),
                 alpenglow_last_voted.clone(),
+                consensus_rewards,
             );
             BLSSigverifyService::new(bls_packet_receiver, verifier)
         };

core/src/tvu.rs

@@ -32,6 +32,7 @@ use {
     },
     anyhow::{anyhow, Context, Result},
     crossbeam_channel::{bounded, unbounded, Receiver},
+    parking_lot::RwLock as PLRwLock,
     quinn::Endpoint,
     solana_accounts_db::{
         accounts_db::{AccountsDbConfig, ACCOUNTS_DB_CONFIG_FOR_TESTING},
@@ -142,6 +143,7 @@ use {
     solana_validator_exit::Exit,
     solana_vote_program::vote_state,
     solana_votor::{
+        consensus_rewards::ConsensusRewards,
         vote_history::{VoteHistory, VoteHistoryError},
         vote_history_storage::{NullVoteHistoryStorage, VoteHistoryStorage},
         voting_service::VotingServiceOverride,
@@ -1425,6 +1427,11 @@ impl Validator {
             migration_status.clone(),
         );
 
+        let consensus_rewards = Arc::new(PLRwLock::new(ConsensusRewards::new(
+            cluster_info.clone(),
+            leader_schedule_cache.clone(),
+        )));
+
         let block_creation_loop_config = BlockCreationLoopConfig {
             exit: exit.clone(),
             migration_status: migration_status.clone(),
@@ -1439,6 +1446,7 @@ impl Validator {
             record_receiver: record_receiver.clone(),
             leader_window_notifier: leader_window_notifier.clone(),
             replay_highest_frozen: replay_highest_frozen.clone(),
+            consensus_rewards: consensus_rewards.clone(),
         };
         let block_creation_loop = BlockCreationLoop::new(block_creation_loop_config);
 
@@ -1693,6 +1701,7 @@ impl Validator {
             key_notifiers.clone(),
             alpenglow_last_voted.clone(),
             migration_status.clone(),
+            consensus_rewards,
         )
         .map_err(ValidatorError::Other)?;
 

core/src/validator.rs

@@ -39,6 +39,7 @@ solana-runtime-transaction = { workspace = true }
 solana-sha256-hasher = { workspace = true }
 solana-transaction = { workspace = true }
 solana-transaction-error = { workspace = true }
+solana-votor-messages = { workspace = true }
 thiserror = { workspace = true }
 
 [dev-dependencies]