A0-1635: Limit nonfinalized block production (#769)

.github/workflows/e2e-tests-main-devnet.yml

@@ -454,7 +454,7 @@ jobs:
       run-e2e-rewards-stake-change,
       run-e2e-rewards-change-stake-force-new-era,
       run-e2e-rewards-points-basic,
-      run-e2e-authorities-are-staking,
+#      run-e2e-authorities-are-staking,
       run-e2e-ban-automatic,
       run-e2e-version-upgrade,
     ]

bin/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "aleph-node"
-version = "0.8.1"
+version = "0.8.2"
 authors = ["Cardinal Cryptography"]
 description = "Aleph node binary"
 edition = "2021"
@@ -43,6 +43,8 @@ sc-network = { git = "https://github.com/Cardinal-Cryptography/substrate.git", b
 sc-transaction-pool = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
 sp-transaction-pool = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
 sc-transaction-pool-api = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
+sp-arithmetic = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
+sc-consensus-slots = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
 sc-consensus-aura = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
 sp-consensus-aura = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }
 sp-consensus = { git = "https://github.com/Cardinal-Cryptography/substrate.git", branch = "aleph-v0.9.28" }

bin/node/src/aleph_cli.rs

@@ -3,6 +3,7 @@ use std::path::PathBuf;
 use aleph_primitives::DEFAULT_UNIT_CREATION_DELAY;
 use clap::{ArgGroup, Parser};
 use finality_aleph::UnitCreationDelay;
+use log::warn;
 
 #[derive(Debug, Parser, Clone)]
 #[clap(group(ArgGroup::new("backup")))]
@@ -33,6 +34,12 @@ pub struct AlephCli {
     /// with `--no-backup`, but note that that limits crash recoverability.
     #[clap(long, value_name = "PATH", group = "backup")]
     backup_path: Option<PathBuf>,
+
+    /// The maximum number of nonfinalized blocks, after which block production should be locally
+    /// stopped. DO NOT CHANGE THIS, PRODUCING MORE OR FEWER BLOCKS MIGHT BE CONSIDERED MALICIOUS
+    /// BEHAVIOUR AND PUNISHED ACCORDINGLY!
+    #[clap(long, default_value_t = 20)]
+    max_nonfinalized_blocks: u32,
 }
 
 impl AlephCli {
@@ -55,4 +62,11 @@ impl AlephCli {
     pub fn no_backup(&self) -> bool {
         self.no_backup
     }
+
+    pub fn max_nonfinalized_blocks(&self) -> u32 {
+        if self.max_nonfinalized_blocks != 20 {
+            warn!("Running block production with a value of max-nonfinalized-blocks {}, which is not the default of 20. THIS MIGHT BE CONSIDERED MALICIOUS BEHAVIOUR AND RESULT IN PENALTIES!", self.max_nonfinalized_blocks);
+        }
+        self.max_nonfinalized_blocks
+    }
 }

bin/node/src/service.rs

@@ -15,15 +15,17 @@ use futures::channel::mpsc;
 use log::warn;
 use sc_client_api::{Backend, ExecutorProvider, HeaderBackend};
 use sc_consensus_aura::{ImportQueueParams, SlotProportion, StartAuraParams};
+use sc_consensus_slots::BackoffAuthoringBlocksStrategy;
 use sc_network::NetworkService;
 use sc_service::{
     error::Error as ServiceError, Configuration, KeystoreContainer, NetworkStarter, RpcHandlers,
     TFullClient, TaskManager,
 };
 use sc_telemetry::{Telemetry, TelemetryWorker};
 use sp_api::ProvideRuntimeApi;
+use sp_arithmetic::traits::BaseArithmetic;
 use sp_blockchain::Backend as _;
-use sp_consensus_aura::sr25519::AuthorityPair as AuraPair;
+use sp_consensus_aura::{sr25519::AuthorityPair as AuraPair, Slot};
 use sp_runtime::{
     generic::BlockId,
     traits::{Block as BlockT, Header as HeaderT, Zero},
@@ -35,6 +37,30 @@ type FullClient = sc_service::TFullClient<Block, RuntimeApi, AlephExecutor>;
 type FullBackend = sc_service::TFullBackend<Block>;
 type FullSelectChain = sc_consensus::LongestChain<FullBackend, Block>;
 
+struct LimitNonfinalized(u32);
+
+impl<N: BaseArithmetic> BackoffAuthoringBlocksStrategy<N> for LimitNonfinalized {
+    fn should_backoff(
+        &self,
+        chain_head_number: N,
+        _chain_head_slot: Slot,
+        finalized_number: N,
+        _slow_now: Slot,
+        _logging_target: &str,
+    ) -> bool {
+        let nonfinalized_blocks: u32 = chain_head_number
+            .saturating_sub(finalized_number)
+            .unique_saturated_into();
+        match nonfinalized_blocks >= self.0 {
+            true => {
+                warn!("We have {} nonfinalized blocks, with the limit being {}, delaying block production.", nonfinalized_blocks, self.0);
+                true
+            }
+            false => false,
+        }
+    }
+}
+
 fn backup_path(aleph_config: &AlephCli, base_path: &Path) -> Option<PathBuf> {
     if aleph_config.no_backup() {
         return None;
@@ -289,7 +315,7 @@ pub fn new_authority(
     );
 
     let force_authoring = config.force_authoring;
-    let backoff_authoring_blocks: Option<()> = None;
+    let backoff_authoring_blocks = Some(LimitNonfinalized(aleph_config.max_nonfinalized_blocks()));
     let prometheus_registry = config.prometheus_registry().cloned();
 
     let (_rpc_handlers, network, network_starter) = setup(