polyfill-rs/README.md

A high-performance Polymarket Rust client with latency-optimized data structures and zero-allocation hot paths. The 0.4.x line is V2-native and intentionally breaking for authenticated trading flows.

At the time that this project was started, polymarket-rs-client was a Polymarket Rust Client with a few GitHub stars, but which seemed to be unmaintained. I took on the task of creating a Rust client which could beat the benchmarks quoted in the README.md of that project, with the added constraint of also maintaining zero alloc hot paths.

I also want to take a moment to clarify what zero-alloc means because I've now recieved double digit messages about this on twitter/x and telegram. In general, zero alloc means either zero alloc in hot paths (which can be a bit more arbitrary) or atlernatively it can mean zero alloc after init/warm-up, which is the objective of this repository. Succinctly that means that the per-message handling loop never touches the heap.

Notably order book paths that introduce new allocations by design:

• First time seeing a token/book (HashMap insert + key clone): src/book.rs:~788
• New price levels (BTreeMap node growth): src/book.rs:~409

Quick Start

Add to your Cargo.toml:

[dependencies]
polyfill-rs = "0.4.0"

Replace your imports:

// Before: use polymarket_rs_client::{ClobClient, Side, OrderType};
use polyfill_rs::{ClobClient, Side, OrderType};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = ClobClient::new("https://clob.polymarket.com");
    let markets = client.get_sampling_markets(None).await?;
    println!("Found {} markets", markets.data.len());
    Ok(())
}

Performance Comparison

Real-World API Performance (with network I/O)

Real-world Polymarket API latency broken down by request phase:

Operation	Metric	polyfill-rs	rs-clob-client-v2	polymarket-rs-client	Official Python Client
Fetch Markets	mean ± sd	321.6 ms ± 92.9 ms	-	409.3 ms ± 137.6 ms	1.366 s ± 0.048 s
Cold Start	single run	651.7 ms	543.9 ms	-	-
Warm Connection	single run	202.9 ms	497.2 ms	-	-
Steady Typed Total	p50 / p95 / p99	209.9 / 276.1 / 509.5 ms	215.9 / 284.7 / 312.4 ms	-	-
Network-Only Byte Fetch	p50 / p95 / p99	382.5 / 590.3 / 626.7 ms	300.1 / 449.0 / 520.1 ms	-	-
CPU Parse Only	p50 / p95 / p99	0.5 / 1.2 / 1.4 ms	1.3 / 1.4 / 1.4 ms	-	-

Performance vs polymarket-rs-client:

• 21.4% faster
• 32.5% more consistent
• 4.2x faster than Official Python Client

Benchmark Methodology: The rs-clob-client-v2 comparison separates cold start, warm connection, steady-state typed requests, network-only byte fetches, and CPU-only parsing. Steady-state rows use 20 paired iterations with alternating order and 100ms delay; parse rows use 200 iterations from a cached 480KB payload. The network-only row compares byte fetches through each HTTP stack without typed deserialization. Run it with cargo run --release --example official_client_side_by_side_benchmark --features official-client-benchmark. See examples/side_by_side_benchmark.rs in commit a63a170: https://github.com/floor-licker/polyfill-rs/blob/a63a170/examples/side_by_side_benchmark.rs for the original legacy benchmark implementation.

Computational Performance (pure CPU, no I/O)

Operation	Performance	Notes
Order Book Updates (1000 ops)	159.6 µs ± 32 µs	6,260 updates/sec, zero-allocation
Spread/Mid Calculations	70 ns ± 77 ns	14.3M ops/sec, optimized BTreeMap
JSON Parsing (480KB)	~2.3 ms	SIMD-accelerated parsing (1.77x faster than serde_json)
WS book hot path (decode + apply)	~0.23 µs / 1.73 µs / 6.74 µs	1 / 16 / 64 levels-per-side, strict fixed-point tape parser with generation-marked snapshot retention (see benches/ws_hot_path.rs)

Run the WS hot-path benchmark locally with cargo bench --bench ws_hot_path.

Key Performance Optimizations:

The 21.4% performance improvement comes from SIMD-accelerated JSON parsing (1.77x faster than serde_json), HTTP/2 tuning with 512KB stream windows optimized for 469KB payloads, integrated DNS caching, connection keep-alive, and buffer pooling to reduce allocation overhead.

Memory Architecture

Pre-allocated pools eliminate allocation latency spikes. Configurable book depth limiting prevents memory bloat. Hot data structures group frequently-accessed fields for cache line efficiency.

Architectural Principles

Price data converts to fixed-point at ingress boundaries while maintaining tick-aligned precision. The critical path uses integer arithmetic with branchless operations. Data converts back to IEEE 754 at egress for API compatibility. This enables deterministic execution with predictable instruction counts.

Measured Network Improvements

Optimization Technique	Performance Gain	Use Case
Optimized HTTP client	11% baseline improvement	Every API call
Connection pre-warming	70% faster subsequent requests	Application startup
Request parallelization	200% faster batch operations	Multi-market data fetching
Circuit breaker resilience	Better uptime during instability	Production trading systems