Trading bot logging and monitoring: see every decision live
A trading bot runs unattended with real money, so when something goes wrong you need to know immediately and be able to reconstruct exactly what happened. That is what logging and monitoring deliver. Logging records every decision the bot makes; monitoring watches the running bot and alerts you the moment it stalls, errors or drifts off plan. Without them, a bot that silently stops trading or starts misbehaving can quietly cost you for days. This guide covers what to log, structured logging, heartbeats, alerting and a dead-man’s switch — with code.
Why logging and monitoring matter
A bot is a black box unless you instrument it. When a trade goes wrong, “what did the bot see and decide?” is unanswerable without logs. And because a bot runs 24/7 unattended, a silent failure — a crashed process, a dead feed, a stuck position — can bleed money for days unless monitoring alerts you. Together they turn an opaque, risky process into one you can trust and debug, complementing solid error handling.
What to log
Log every meaningful event: each signal computed (with the indicator values that produced it), every order sent and its fill, every error and retry, balance and position snapshots, and a periodic heartbeat. The rule is simple — if you would need it to explain a surprising trade after the fact, log it.
Structured logging
python · logger.pyimport logging, json
logging.basicConfig(filename='bot.log', level=logging.INFO)
log = logging.getLogger('bot')
def event(kind, **data):
log.info(json.dumps({'kind': kind, **data}))
event('signal', symbol='BTC/USDT', rsi=28.4, action='buy')
event('order', side='buy', qty=0.01, price=60150, id='abc')
JSON lines are machine-parseable, so you can later filter, chart or feed them into a log service — far better than free-text prints.
Heartbeat and alerting
python · alert.pyimport requests
def alert(msg):
requests.post(f"https://api.telegram.org/bot{TOKEN}/sendMessage",
data={'chat_id': CHAT, 'text': msg})
# call every loop; if no heartbeat in N minutes, you know it died
alert("⚠️ order rejected: insufficient funds")
A Telegram or email alert on errors and a regular heartbeat means you hear about a problem in minutes, not days.
Dashboards and metrics
For a serious setup, ship metrics (equity, open positions, daily P&L, error rate) to a dashboard so you can glance at health. Even a simple daily summary message beats nothing. Track the same risk metrics your backtester reports — live drawdown versus expected is the number that tells you the bot has gone off the rails.
The dead-man’s switch
Combine monitoring with action: if the heartbeat stops, or live drawdown breaches your limit, automatically halt the bot and flatten positions. Monitoring that only watches is half a system; monitoring that also acts is the safety net. This ties directly into the kill switch from error handling and good deployment practice.
Frequently asked questions
What should a trading bot log?
A trading bot should log every meaningful event: each signal it computes along with the indicator values behind it, every order sent and its fill, every error and retry, periodic balance and position snapshots, and a regular heartbeat. The guiding rule is to log anything you would need to explain a surprising trade after the fact, because a black-box bot is impossible to debug.
How do I monitor a trading bot for failures?
Combine a heartbeat — a regular signal the bot is alive — with alerting that messages you (via Telegram or email) on errors or when the heartbeat stops. A monitor watching the heartbeat tells you within minutes if the bot crashes, a feed dies or a position gets stuck, instead of letting a silent failure bleed money for days while you assume it is fine.
What is structured logging in a trading bot?
Structured logging records events as machine-parseable data, typically JSON lines containing fields like event kind, symbol, indicator values, order side, quantity and price, rather than free-text print statements. Because each entry is structured, you can later filter, chart or feed the logs into a log service to reconstruct exactly what the bot saw and decided.
What is a dead-man’s switch for a trading bot?
A dead-man’s switch automatically halts the bot and flattens positions when something goes wrong — for example if the heartbeat stops or live drawdown breaches your limit. It turns passive monitoring into active protection, so a crashed process or a runaway loss triggers a safe shutdown rather than continuing to trade unattended and compounding the damage.