Release v3.6.5: SWQOS core affinity and recommended sender thread indices

- Add TradeConfig::with_swqos_cores_from_end(bool) to use last N CPU cores for SWQOS, reducing contention with main thread and default tokio workers.
- Add recommended_sender_thread_core_indices(swqos_count) to get the same last-N core indices for with_dedicated_sender_threads (recommended combo for lower latency).
- Document core affinity and latency in async_executor and with_dedicated_sender_threads.

Made-with: Cursor

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "sol-trade-sdk"
-version = "3.6.4"
+version = "3.6.5"
 edition = "2021"
 authors = [
     "William <byteblock6@gmail.com>",

src/common/types.rs

@@ -9,6 +9,8 @@ pub struct InfrastructureConfig {
     pub rpc_url: String,
     pub swqos_configs: Vec<SwqosConfig>,
     pub commitment: CommitmentConfig,
+    /// When true, SWQOS sender threads use the *last* N cores instead of the first N. Reduces contention with main thread / default tokio workers that often use low-numbered cores. Default false.
+    pub swqos_cores_from_end: bool,
 }
 
 impl InfrastructureConfig {
@@ -17,7 +19,12 @@ impl InfrastructureConfig {
         swqos_configs: Vec<SwqosConfig>,
         commitment: CommitmentConfig,
     ) -> Self {
-        Self { rpc_url, swqos_configs, commitment }
+        Self {
+            rpc_url,
+            swqos_configs,
+            commitment,
+            swqos_cores_from_end: false,
+        }
     }
 
     /// Create from TradeConfig (extract infrastructure-only settings)
@@ -26,6 +33,7 @@ impl InfrastructureConfig {
             rpc_url: config.rpc_url.clone(),
             swqos_configs: config.swqos_configs.clone(),
             commitment: config.commitment.clone(),
+            swqos_cores_from_end: config.swqos_cores_from_end,
         }
     }
 
@@ -43,8 +51,8 @@ impl Hash for InfrastructureConfig {
     fn hash<H: Hasher>(&self, state: &mut H) {
         self.rpc_url.hash(state);
         self.swqos_configs.hash(state);
-        // Hash commitment level as string since CommitmentConfig doesn't impl Hash
         format!("{:?}", self.commitment).hash(state);
+        self.swqos_cores_from_end.hash(state);
     }
 }
 
@@ -53,6 +61,7 @@ impl PartialEq for InfrastructureConfig {
         self.rpc_url == other.rpc_url
             && self.swqos_configs == other.swqos_configs
             && self.commitment == other.commitment
+            && self.swqos_cores_from_end == other.swqos_cores_from_end
     }
 }
 
@@ -72,6 +81,8 @@ pub struct TradeConfig {
     pub log_enabled: bool,
     /// Whether to check minimum tip per SWQOS provider (filter out configs below min). Default false to save latency.
     pub check_min_tip: bool,
+    /// When true, SWQOS uses the *last* N cores (instead of the first N). Use when main thread / tokio use low-numbered cores to reduce CPU contention. Default false.
+    pub swqos_cores_from_end: bool,
 }
 
 impl TradeConfig {
@@ -91,7 +102,8 @@ impl TradeConfig {
             create_wsol_ata_on_startup: true, // default: check and create on startup
             use_seed_optimize: true, // default: use seed optimization
             log_enabled: true,      // default: enable all SDK logs
-            check_min_tip: false,             // default: skip min tip check to reduce latency
+            check_min_tip: false,   // default: skip min tip check to reduce latency
+            swqos_cores_from_end: false,
         }
     }
 
@@ -111,6 +123,12 @@ impl TradeConfig {
         self.check_min_tip = check_min_tip;
         self
     }
+
+    /// Use the *last* N cores for SWQOS (instead of the first N). Call this when the main thread or tokio workers use low-numbered cores to avoid binding SWQOS to busy cores. Default false.
+    pub fn with_swqos_cores_from_end(mut self, from_end: bool) -> Self {
+        self.swqos_cores_from_end = from_end;
+        self
+    }
 }
 
 pub type SolanaRpcClient = solana_client::nonblocking::rpc_client::RpcClient;

src/lib.rs

@@ -185,7 +185,15 @@ impl TradingInfrastructure {
             let max_by_cores = (num_cores * 2 / 3).max(1);
             let cap = swqos_count.min(max_by_cores).max(1);
             let ids = core_affinity::get_core_ids()
-                .map(|all| all.into_iter().take(cap).collect::<Vec<_>>())
+                .map(|all| {
+                    let v: Vec<_> = all.into_iter().collect();
+                    let len = v.len();
+                    if config.swqos_cores_from_end && len >= cap {
+                        v.into_iter().skip(len - cap).collect()
+                    } else {
+                        v.into_iter().take(cap).collect()
+                    }
+                })
                 .unwrap_or_default();
             (cap, Arc::new(ids))
         };
@@ -200,6 +208,23 @@ impl TradingInfrastructure {
     }
 }
 
+/// When using `TradeConfig::with_swqos_cores_from_end(true)`, returns the same "last N" core indices
+/// that the infrastructure uses. Pass the result to `TradingClient::with_dedicated_sender_threads`
+/// for 方式 C (组合使用): SWQOS on last N cores and dedicated sender threads pinned to those cores.
+///
+/// Returns `None` if core count cannot be determined. `swqos_count` is typically `swqos_configs.len()`.
+pub fn recommended_sender_thread_core_indices(swqos_count: usize) -> Option<Vec<usize>> {
+    let all = core_affinity::get_core_ids()?;
+    let num_cores = all.len();
+    if num_cores == 0 {
+        return None;
+    }
+    let max_by_cores = (num_cores * 2 / 3).max(1);
+    let cap = swqos_count.min(max_by_cores).max(1).min(num_cores);
+    let start = num_cores.saturating_sub(cap);
+    Some((start..num_cores).collect())
+}
+
 /// Main trading client for Solana DeFi protocols
 ///
 /// `SolTradingSDK` provides a unified interface for trading across multiple Solana DEXs
@@ -636,6 +661,9 @@ impl TradingClient {
     /// - `None`: keep default (shared tokio pool).  
     /// - `Some(vec![])`: dedicated threads with default count, no core pinning.  
     /// - `Some(indices)`: dedicated threads pinned to those core indices (trimmed to cap).  
+    ///  
+    /// **Latency note:** If a core is busy with other work (node, bot), SWQOS submit on that core can be delayed.  
+    /// For lowest latency, pass core indices that are *reserved* for SWQOS (do not run other CPU-heavy work on those cores).
     pub fn with_dedicated_sender_threads(mut self, core_indices: Option<Vec<usize>>) -> Self {
         match core_indices {
             None => {

src/trading/core/async_executor.rs

@@ -5,6 +5,11 @@
 //! - **Dedicated threads** (opt-in via `with_dedicated_sender_threads`): N OS threads run sender work only, optionally pinned to cores.
 //! - **Arc**: Shared data behind `Arc` → clone = refcount increment (no data copy).
 //! - **Refs**: `build_transaction` takes refs only; worker path avoids extra clones.
+//!
+//! **Core affinity & latency:** Each job is assigned a core (round-robin from `effective_core_ids`). When a worker runs a job,
+//! it sets thread affinity to that core. If that core is busy with other work (e.g. node sync, bot logic), SWQOS submit on that
+//! core will compete for CPU and latency can increase. For lowest latency, reserve a subset of cores for SWQOS only via
+//! `with_dedicated_sender_threads(Some(indices))` and avoid running other CPU-heavy work on those core indices.
 
 use anyhow::{anyhow, Result};
 use crossbeam_queue::ArrayQueue;