Add opt-in mimalloc memory allocator

- Add mimalloc feature to nautilus-pyo3, enabled for Python wheels
- Set mimalloc as the global allocator in the nautilus CLI binary
- Document allocator in README, architecture and Rust guides
- Add cargo-vet exemptions for mimalloc crates
- Resolves #4358

Author: cjdsellers

.supply-chain/config.toml

@@ -1591,6 +1591,10 @@ criteria = "safe-to-deploy"
 version = "0.2.16"
 criteria = "safe-to-deploy"
 
+[[exemptions.libmimalloc-sys]]
+version = "0.1.49"
+criteria = "safe-to-deploy"
+
 [[exemptions.libsqlite3-sys]]
 version = "0.37.0"
 criteria = "safe-to-deploy"
@@ -1655,6 +1659,10 @@ criteria = "safe-to-deploy"
 version = "2.8.2"
 criteria = "safe-to-deploy"
 
+[[exemptions.mimalloc]]
+version = "0.1.52"
+criteria = "safe-to-deploy"
+
 [[exemptions.mime]]
 version = "0.3.17"
 criteria = "safe-to-deploy"

Cargo.lock

@@ -187,6 +187,7 @@ log = { version = "0.4.33", features = [
   "release_max_level_debug",
 ] }
 madsim = "0.2.34"
+mimalloc = "0.1.52"
 object_store = { version = "0.13.2", default-features = false, features = [
   "fs",
 ] }

Cargo.toml

@@ -51,7 +51,7 @@ DEX), traditional markets (FX, equities, futures, options), and betting exchange
 
 ## Features
 
-- **Fast**: Rust core with asynchronous networking using [tokio](https://crates.io/crates/tokio).
+- **Fast**: Rust core with the [mimalloc](https://github.com/microsoft/mimalloc) allocator and asynchronous networking using [tokio](https://crates.io/crates/tokio).
 - **Reliable**: Type- and thread-safety backed by Rust, with optional Redis-backed state persistence.
 - **Portable**: Runs on Linux, macOS, and Windows. Deploy using Docker.
 - **Flexible**: Modular adapters integrate any REST API or WebSocket feed.

README.md

@@ -24,6 +24,7 @@ installation. Follow-up `2.0.0rcN` wheels are likely to ship at a higher cadence
 releases as feedback arrives, before the final `2.0.0` release.
 
 ### Enhancements
+- Added opt-in `mimalloc` allocator feature, enabled by default for Python wheels (#4358), thanks @ivannp
 - Added `LiveNode` metrics for Rust live runner metrics
 - Added returns skewness and kurtosis portfolio statistics (#4334), thanks @Martingale42
 - Added WebSocket transport backend selection for Python and PyO3 configs (#4342), thanks @graceyangfan

RELEASES.md

@@ -38,6 +38,7 @@ anyhow = { workspace = true }
 clap = { workspace = true }
 dotenvy = { workspace = true }
 log = { workspace = true }
+mimalloc = { workspace = true }
 serde_json = { workspace = true }
 tokio = { workspace = true }
 tokio-util = { workspace = true }

crates/cli/Cargo.toml

@@ -18,9 +18,13 @@
 use std::process::ExitCode;
 
 use clap::FromArgMatches;
+use mimalloc::MiMalloc;
 use nautilus_cli::opt::NautilusCli;
 use nautilus_common::logging::ensure_logging_initialized;
 
+#[global_allocator]
+static GLOBAL: MiMalloc = MiMalloc;
+
 #[tokio::main]
 async fn main() -> ExitCode {
     dotenvy::dotenv().ok();

crates/cli/src/bin/cli.rs

@@ -101,6 +101,7 @@ high-precision = [
   "nautilus-tardis/high-precision",
 ]
 hypersync = ["defi", "nautilus-blockchain/hypersync"]
+mimalloc = ["dep:mimalloc"]
 postgres = ["nautilus-infrastructure/postgres"]
 redis = ["nautilus-infrastructure/redis"]
 tracing-bridge = ["nautilus-common/tracing-bridge"]
@@ -159,6 +160,10 @@ nautilus-polymarket = { workspace = true, features = ["python"] }
 nautilus-sandbox = { workspace = true, features = ["python"] }
 nautilus-tardis = { workspace = true, features = ["python", "replay"] }
 
+# local_dynamic_tls avoids exhausting static TLS space when the extension is dlopen'd
+mimalloc = { workspace = true, features = [
+  "local_dynamic_tls",
+], optional = true }
 pyo3 = { workspace = true }
 pyo3-stub-gen = { workspace = true }
 

crates/pyo3/Cargo.toml

@@ -41,6 +41,7 @@
 //! - `hypersync`: Enables hypersync support (fast parallel hash maps) where available.
 //! - `tracing-bridge`: Enables the `tracing` subscriber bridge for log integration.
 //! - `defi`: Enables DeFi (Decentralized Finance) support including blockchain adapters.
+//! - `mimalloc`: Sets [mimalloc](https://github.com/microsoft/mimalloc) as Rust's global allocator (enabled for binary wheels).
 
 #![warn(rustc::all)]
 #![deny(unsafe_code)]
@@ -53,9 +54,15 @@
 
 use std::{path::Path, time::Duration};
 
+#[cfg(feature = "mimalloc")]
+use mimalloc::MiMalloc;
 use nautilus_common::live::runtime::shutdown_runtime;
 use pyo3::{prelude::*, pyfunction};
 
+#[cfg(feature = "mimalloc")]
+#[global_allocator]
+static GLOBAL: MiMalloc = MiMalloc;
+
 const RUNTIME_SHUTDOWN_TIMEOUT_SECS: u64 = 10;
 
 #[pyfunction]

crates/pyo3/src/lib.rs

@@ -708,6 +708,23 @@ For production deployments, add multiple strategies to a **single TradingNode**
 For parallel execution or workload isolation, run each node in its own separate process.
 :::
 
+### Memory allocation
+
+The event-driven core allocates and frees small objects at high frequency: message bus dispatch,
+order event handling, and order book maintenance all exercise the heap on every event. Default
+system allocators handle this pattern poorly; profiling shows allocator overhead approaching half
+of hot-loop time on both the Windows CRT heap and glibc malloc under order-flow workloads.
+
+The Python wheels and the `nautilus` CLI binary therefore use
+[mimalloc](https://github.com/microsoft/mimalloc) for Rust allocations.
+Backtest engine benchmarks run roughly 3% to 44% faster depending on workload,
+with order-flow heavy paths gaining the most. The trade-off is a modest increase in resident
+memory from mimalloc's segment caching.
+
+A Rust binary links exactly one global allocator, and libraries do not impose one, so the
+NautilusTrader crates remain allocator-neutral. When building directly against the crates,
+opt in from your own binary (see the [Rust guide](rust.md#memory-allocator)).
+
 ## Related guides
 
 - [Overview](overview.md) - High-level introduction to NautilusTrader.