01_exchange

Prompt file imported from acid0ikario/trade-bot (.codex/prompts/01_exchange.md). Copyright stays with the author.

You are acting as a senior Python trading infra engineer. Our project already includes a partially implemented Exchange wrapper in src/bot/exchange.py. Your job is to review and refine this wrapper to ensure it is robust, safe, and fully aligned with our requirements for Binance spot trading.

Context

• This bot trades crypto on Binance spot only; no withdrawals are allowed.
• The wrapper must load API credentials from the environment via EnvVars, enforce a symbol whitelist, and respect a max notional per trade (max_notional_per_trade_usdt) defined in config.yaml.
• The current implementation provides methods like get_price, fetch_ohlcv, get_balance, create_market_buy, create_market_sell, and place_oco_takeprofit_stoploss, along with helper functions for quantity and price rounding.

Goals

1. Audit the existing code in src/bot/exchange.py to identify missing logic or edge cases.
2. Harden safety features: enforce whitelist and notional caps, apply Binance lot-size and min-notional filters, and ensure no withdrawal functionality.
3. Improve reliability: implement exponential backoff for retryable ccxt errors (NetworkError, DDoSProtection, RateLimitExceeded, etc.), and wrap all failures in a custom ExchangeError.
4. Provide comprehensive tests that mock ccxt interactions and cover both happy and failure paths.

Tasks

1. Review and refine the Exchange class:

   • Ensure API keys from EnvVars are used and sanitized (remove None values).
   • Confirm that symbols_whitelist and max_notional_per_trade_usdt in the loaded config are actively checked in every order-related method.
   • Verify that load_markets() is called (or retried) to obtain market filters, and add helper methods to parse LOT_SIZE, PRICE_FILTER, and MIN_NOTIONAL fields.
   • Revisit _prepare_order_qty() to floor quantities to step size, respect minimum quantity and notional constraints, and throw ExchangeError on invalid input (e.g., zero quantity, notional too high).

2. Implement or enhance public methods:

   • get_price(symbol): fetch the last/close price from fetch_ticker, raise ExchangeError if price is missing, and wrap in retry logic.
   • fetch_ohlcv(symbol, timeframe, limit): fetch candles and return as a list of [timestamp, open, high, low, close, volume]; wrap in retry logic.
   • get_balance(quote): return the free balance of a currency; handle both top-level and nested dictionary formats returned by ccxt.
   • create_market_buy / create_market_sell: compute adjusted quantity via _prepare_order_qty(), log the order, and place a market order with retries.
   • place_oco_takeprofit_stoploss: simulate an OCO order by placing a limit sell at TP and a stop-limit sell at SL; round TP/SL to the nearest tick size, adjust quantity, and return both order IDs.

3. Safety and reliability:

   • Use an internal _with_retries() helper implementing exponential backoff (e.g., base delay 0.05 s doubling up to 1 s) for transient ccxt errors, and raise ExchangeError on non-retryable errors.
   • Normalize all ccxt exceptions into our custom ExchangeError.
   • Do not include any withdrawal methods in this wrapper.

4. Unit tests (tests/test_exchange.py):

   • Use pytest and monkeypatch to simulate ccxt responses without hitting the network.
   • Test success cases: price retrieval, OHLCV retrieval, balance retrieval, successful buys, sells, and OCO orders.
   • Test failure cases: symbol not whitelisted, quantity below minimum, notional cap exceeded, missing price in ticker, and correct handling of step-size flooring.
   • Test retry logic by having the mock raise a retryable error once before succeeding.

5. Acceptance criteria:

   • All methods in Exchange behave as specified and raise ExchangeError on invalid inputs or API errors.
   • Tests in tests/test_exchange.py pass locally (run with poetry run pytest -q).
   • The wrapper enforces whitelist and notional caps, adjusts quantities for market filters, uses exponential backoff on transient errors, and logs order actions via loguru.