Rate Limiting

Three-Layer Architecture

Requests pass through three independent layers in order. All layers must allow a request before it is dispatched.

Circuit Breaker State Machine

The global limiter includes a circuit breaker that automatically suspends requests when error rates spike, then probes recovery before resuming full traffic.

Client-Side Rate Limit Groups

Binance enforces separate weight limits per API group. AutoP2P tracks usage per group using response headers.

Weight Tracking

After every response, AutoP2P reads the x-mbx-used-weight-1m header and compares it to the known limit. When usage exceeds WEIGHT_BACKOFF_THRESHOLD (default 80%), the global limiter slows token emission to avoid hitting the hard ceiling.

# Pseudo-logic in global_limiter.on_response()
used = int(response.headers.get("x-mbx-used-weight-1m", 0))
ratio = used / 12_000  # WEIGHT_LIMIT

if ratio >= WEIGHT_BACKOFF_THRESHOLD:  # default 0.8
    # Reduce token emission rate proportionally
    limiter.set_rate(GLOBAL_BINANCE_RPS * (1 - ratio))
else:
    limiter.set_rate(GLOBAL_BINANCE_RPS)

Empty response arrays from the public BAPI search (EP-0) trigger a separate soft rate limit detector. After EMPTY_RESPONSE_THRESHOLD consecutive empty results, the limiter adds a cooldown before the next search request.

Best Practices

All Environment Variables

All rate limiting behaviour is configurable via environment variables. Defaults are tuned for safe production operation with a single Binance account.