feat(host-listener): throttle dependent ops

Author: Eikix

coprocessor/fhevm-engine/host-listener/src/bin/poller.rs

@@ -118,6 +118,20 @@ struct Args {
         help = "Dependence chain are across blocks"
     )]
     pub dependence_cross_block: bool,
+
+    #[arg(
+        long,
+        default_value_t = 0,
+        help = "Global dependent ops rate limit per minute (0 disables)"
+    )]
+    pub dependent_ops_rate_per_min: u32,
+
+    #[arg(
+        long,
+        default_value_t = 0,
+        help = "Burst size for dependent ops limiter (0 = same as rate)"
+    )]
+    pub dependent_ops_burst: u32,
 }
 
 #[tokio::main]
@@ -159,6 +173,8 @@ async fn main() -> anyhow::Result<()> {
         dependence_cache_size: args.dependence_cache_size,
         dependence_by_connexity: args.dependence_by_connexity,
         dependence_cross_block: args.dependence_cross_block,
+        dependent_ops_rate_per_min: args.dependent_ops_rate_per_min,
+        dependent_ops_burst: args.dependent_ops_burst,
     };
 
     run_poller(config).await

coprocessor/fhevm-engine/host-listener/src/cmd/mod.rs

@@ -125,6 +125,20 @@ pub struct Args {
     )]
     pub dependence_cross_block: bool,
 
+    #[arg(
+        long,
+        default_value_t = 0,
+        help = "Global dependent ops rate limit per minute (0 disables)"
+    )]
+    pub dependent_ops_rate_per_min: u32,
+
+    #[arg(
+        long,
+        default_value_t = 0,
+        help = "Burst size for dependent ops limiter (0 = same as rate)"
+    )]
+    pub dependent_ops_burst: u32,
+
     #[arg(
         long,
         default_value = "50",
@@ -1047,6 +1061,10 @@ pub async fn main(args: Args) -> anyhow::Result<()> {
         args.dependence_cache_size,
     )
     .await?;
+    db.set_dependent_ops_limiter(
+        args.dependent_ops_rate_per_min,
+        args.dependent_ops_burst,
+    );
 
     if chain_id != db.chain_id {
         error!(

coprocessor/fhevm-engine/host-listener/src/database/tfhe_event_propagate.rs

@@ -9,15 +9,19 @@ use fhevm_engine_common::types::AllowEvents;
 use fhevm_engine_common::types::SupportedFheOperations;
 use fhevm_engine_common::utils::DatabaseURL;
 use fhevm_engine_common::utils::{to_hex, HeartBeat};
+use prometheus::{register_int_counter_vec, IntCounterVec};
 use sqlx::postgres::PgConnectOptions;
 use sqlx::postgres::PgPoolOptions;
 use sqlx::types::Uuid;
 use sqlx::Error as SqlxError;
 use sqlx::{PgPool, Postgres};
 use std::ops::DerefMut;
 use std::sync::Arc;
+use std::sync::LazyLock;
 use std::time::Duration;
+use std::time::Instant;
 use time::{Duration as TimeDuration, PrimitiveDateTime};
+use tokio::sync::Mutex;
 use tokio::sync::RwLock;
 use tracing::error;
 use tracing::info;
@@ -39,6 +43,24 @@ pub type ScalarByte = FixedBytes<1>;
 pub type ClearConst = Uint<256, 4>;
 pub type ChainHash = TransactionHash;
 
+static DEPENDENT_OPS_ALLOWED: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    register_int_counter_vec!(
+        "host_listener_dependent_ops_allowed",
+        "Number of dependent ops allowed by the limiter",
+        &["chain_id"]
+    )
+    .unwrap()
+});
+
+static DEPENDENT_OPS_THROTTLED: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    register_int_counter_vec!(
+        "host_listener_dependent_ops_throttled",
+        "Number of dependent ops deferred by the limiter",
+        &["chain_id"]
+    )
+    .unwrap()
+});
+
 #[derive(Clone, Debug)]
 pub struct Chain {
     pub hash: ChainHash,
@@ -92,8 +114,10 @@ pub struct Database {
     pub pool: Arc<RwLock<sqlx::Pool<Postgres>>>,
     pub tenant_id: TenantId,
     pub chain_id: ChainId,
+    chain_id_label: String,
     pub dependence_chain: ChainCache,
     pub tick: HeartBeat,
+    dependent_ops_limiter: Option<Arc<Mutex<DependentOpsLimiter>>>,
 }
 
 #[derive(Debug)]
@@ -107,6 +131,67 @@ pub struct LogTfhe {
     pub dependence_chain: TransactionHash,
 }
 
+#[derive(Debug)]
+struct DependentOpsLimiter {
+    rate_per_min: u32,
+    burst: u32,
+    tokens: f64,
+    last_refill: Instant,
+}
+
+impl DependentOpsLimiter {
+    fn new(rate_per_min: u32, burst: u32) -> Option<Self> {
+        if rate_per_min == 0 {
+            return None;
+        }
+        let burst = if burst == 0 {
+            rate_per_min.max(1)
+        } else {
+            burst
+        };
+        Some(Self {
+            rate_per_min,
+            burst,
+            tokens: burst as f64,
+            last_refill: Instant::now(),
+        })
+    }
+
+    fn defer_duration(&mut self) -> Duration {
+        let now = Instant::now();
+        self.refill(now);
+        let rate_per_sec = self.rate_per_min as f64 / 60.0;
+        if rate_per_sec <= 0.0 {
+            return Duration::ZERO;
+        }
+        self.tokens -= 1.0;
+        if self.tokens >= 0.0 {
+            Duration::ZERO
+        } else {
+            let deficit = -self.tokens;
+            let wait_secs = deficit / rate_per_sec;
+            if wait_secs <= 0.0 {
+                Duration::ZERO
+            } else {
+                Duration::from_secs_f64(wait_secs)
+            }
+        }
+    }
+
+    fn refill(&mut self, now: Instant) {
+        let elapsed = now.duration_since(self.last_refill).as_secs_f64();
+        if elapsed <= 0.0 {
+            return;
+        }
+        let rate_per_sec = self.rate_per_min as f64 / 60.0;
+        let added = elapsed * rate_per_sec;
+        if added > 0.0 {
+            self.tokens = (self.tokens + added).min(self.burst as f64);
+            self.last_refill = now;
+        }
+    }
+}
+
 pub type Transaction<'l> = sqlx::Transaction<'l, Postgres>;
 
 impl Database {
@@ -129,12 +214,20 @@ impl Database {
             url: url.clone(),
             tenant_id,
             chain_id,
+            chain_id_label: chain_id.to_string(),
             pool: Arc::new(RwLock::new(pool)),
             dependence_chain: bucket_cache,
             tick: HeartBeat::default(),
+            dependent_ops_limiter: None,
         })
     }
 
+    pub fn set_dependent_ops_limiter(&mut self, rate_per_min: u32, burst: u32) {
+        self.dependent_ops_limiter =
+            DependentOpsLimiter::new(rate_per_min, burst)
+                .map(|limiter| Arc::new(Mutex::new(limiter)));
+    }
+
     async fn new_pool(url: &DatabaseURL) -> PgPool {
         let options: PgConnectOptions = url.parse().expect("bad url");
         let options = options.options([
@@ -280,6 +373,32 @@ impl Database {
     ) -> Result<bool, SqlxError> {
         let is_scalar = !scalar_byte.is_zero();
         let output_handle = result.to_vec();
+        let mut schedule_order =
+            log.block_timestamp
+                .saturating_add(TimeDuration::microseconds(
+                    log.tx_depth_size as i64,
+                ));
+        if log.is_allowed && !dependencies.is_empty() {
+            if let Some(limiter) = &self.dependent_ops_limiter {
+                let defer_for = limiter.lock().await.defer_duration();
+                if defer_for.is_zero() {
+                    DEPENDENT_OPS_ALLOWED
+                        .with_label_values(&[self.chain_id_label.as_str()])
+                        .inc();
+                } else {
+                    // Defer by writing a future schedule_order; worker still
+                    // pulls earliest schedule_order, so this smooths load without blocking ingest.
+                    let defer_micros =
+                        defer_for.as_micros().min(i64::MAX as u128) as i64;
+                    schedule_order = schedule_order.saturating_add(
+                        TimeDuration::microseconds(defer_micros),
+                    );
+                    DEPENDENT_OPS_THROTTLED
+                        .with_label_values(&[self.chain_id_label.as_str()])
+                        .inc();
+                }
+            }
+        }
         let query = sqlx::query!(
             r#"
             INSERT INTO computations (
@@ -306,10 +425,7 @@ impl Database {
             log.dependence_chain.to_vec(),
             log.transaction_hash.map(|txh| txh.to_vec()),
             log.is_allowed,
-            log.block_timestamp
-                .saturating_add(time::Duration::microseconds(
-                    log.tx_depth_size as i64
-                )),
+            schedule_order,
             !log.is_allowed,
         );
         query
@@ -1094,3 +1210,34 @@ pub fn tfhe_inputs_handle(op: &TfheContractEvents) -> Vec<Handle> {
         E::Initialized(_) | E::Upgraded(_) | E::VerifyInput(_) => vec![],
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn dependent_ops_limiter_defers_after_burst() {
+        let mut limiter = DependentOpsLimiter::new(60, 2).unwrap();
+        assert!(limiter.defer_duration().is_zero());
+        assert!(limiter.defer_duration().is_zero());
+        let deferred = limiter.defer_duration();
+        assert!(deferred > Duration::ZERO);
+    }
+
+    #[test]
+    fn dependent_ops_limiter_refills_over_time() {
+        let mut limiter = DependentOpsLimiter::new(60, 1).unwrap();
+        assert!(limiter.defer_duration().is_zero());
+        let deferred = limiter.defer_duration();
+        assert!(deferred > Duration::ZERO);
+        limiter.last_refill =
+            Instant::now().checked_sub(Duration::from_secs(2)).unwrap();
+        let allowed_after_refill = limiter.defer_duration();
+        assert!(allowed_after_refill.is_zero());
+    }
+
+    #[test]
+    fn dependent_ops_limiter_disabled_when_rate_zero() {
+        assert!(DependentOpsLimiter::new(0, 10).is_none());
+    }
+}

coprocessor/fhevm-engine/host-listener/src/poller/mod.rs

@@ -87,6 +87,8 @@ pub struct PollerConfig {
     pub dependence_cache_size: u16,
     pub dependence_by_connexity: bool,
     pub dependence_cross_block: bool,
+    pub dependent_ops_rate_per_min: u32,
+    pub dependent_ops_burst: u32,
 }
 
 pub async fn run_poller(config: PollerConfig) -> Result<()> {
@@ -139,6 +141,10 @@ pub async fn run_poller(config: PollerConfig) -> Result<()> {
         config.dependence_cache_size,
     )
     .await?;
+    db.set_dependent_ops_limiter(
+        config.dependent_ops_rate_per_min,
+        config.dependent_ops_burst,
+    );
 
     if chain_id != db.chain_id {
         error!(

coprocessor/fhevm-engine/host-listener/tests/host_listener_integration_tests.rs

@@ -234,6 +234,8 @@ async fn setup(node_chain_id: Option<u64>) -> Result<Setup, anyhow::Error> {
         catchup_finalization_in_blocks: 2,
         dependence_by_connexity: false,
         dependence_cross_block: true,
+        dependent_ops_rate_per_min: 0,
+        dependent_ops_burst: 0,
         timeout_request_websocket: 30,
     };
     let health_check_url = format!("http://127.0.0.1:{}", args.health_port);
@@ -289,6 +291,8 @@ async fn test_only_catchup_loop_requires_negative_start_at_block(
         catchup_loop_sleep_secs: 60,
         dependence_by_connexity: false,
         dependence_cross_block: true,
+        dependent_ops_rate_per_min: 0,
+        dependent_ops_burst: 0,
         timeout_request_websocket: 30,
     };
 

coprocessor/fhevm-engine/host-listener/tests/poller_integration_tests.rs

@@ -200,6 +200,8 @@ async fn poller_catches_up_to_safe_tip(
         dependence_cache_size: 10_000,
         dependence_by_connexity: false,
         dependence_cross_block: false,
+        dependent_ops_rate_per_min: 0,
+        dependent_ops_burst: 0,
     };
 
     let poller_handle = tokio::spawn(run_poller(config));