Async Event-Driven Architecture for a Modular Trading Bot
Article OctoBot
Summary
This document explains OctoBot’s architecture for moving market and other data through a modular trading system. It uses asynchronous producer-consumer channels to notify components when new information arrives, aiming to avoid polling loops and keep downstream evaluations current. The design also seeks to limit thread use to reduce CPU overhead, though no benchmark results are provided.
The system separates evaluators, strategies, and trading modes. Evaluators handle focused tasks such as measuring an indicator; strategies combine evaluations across timeframes into decisions; trading modes translate those decisions into order actions while considering funds, open orders, and protective settings. Extensions called tentacles provide evaluation logic and utility services, while channel events trigger the next stage. The document outlines these roles and event flows, but gives no quantitative comparison, latency measurements, or details for assessing how performance changes across workloads.
Key ideas
- Async channels notify components when new data arrives, avoiding periodic polling loops.
- Evaluators perform focused analyses and publish their results for strategies to consume.
- Strategies combine evaluator outputs into higher-level trading decisions.
- Trading modes turn strategy evaluations into order creation, modification, or cancellation.
- The document describes intended efficiency but provides no benchmark evidence.
Tags
Full text
# Design Philosophy --- title: "Design Philosophy" description: "Learn about the OctoBot design philosophy and technical architecture based with speed and scalability in mind using Python and asynchronous programming with asyncio." sidebar_position: 2 --- # Design Philosophy ## Philosophy The goal behind OctoBot is to have a **very fast and scalable** trading robot. To achieve this, OctoBot is entirely built around the <a href="https://docs.python.org/3/library/asyncio.html" rel="nofollow">asyncio</a> producer-consumer <a href="https://github.com/Drakkar-Software/Async-Channel" rel="nofollow">Async-Channel</a> framework which allows to very quickly and efficiently transmit data to different elements within the bot. The idea is to all the time maintain **fully up-to-date data** without having to use update loops. Update loops require sleeping time, which is inefficient. This architecture enables to **notify the evaluation chain as quickly as possible** when an update is available without having to wait for any update cycle of any update loop. Additionally, in order to save CPU time, as little threads as possible are used by OctoBot (usually less than 10 with a standard setup). ## Overview The OctoBot code is split into [several repositories](github-repositories). Each module is handled as an independent python module and is available on the <a href="https://pypi.org/" rel="nofollow">official python package repository</a> (used in `pip` commands). ## OctoBot  Simplified view of the OctoBot core components. Inside the OctoBot part, each arrow is an async channel. ## OctoBot tentacles Tentacles are OctoBot's extensions, they are meant to be easily customizable, can be activated or not and do any specific action within OctoBot. ### Evaluation chain tentacles They are tools to analyze market data as well as any other type of data (Teddit, Telegram, etc). They implement abstract evaluators, strategies and trading modes. ### Utility tentacles These are OctoBot's interfaces (web, telegram), notification systems, social news feeds and [backtesting](/guides/octobot-usage/backtesting) data collectors. They implement abstract interfaces, services, service feeds, notifiers and data collectors ## Evaluators, strategies and trading modes: ### Evaluators Simple python classes that will automatically be wake up when new data is available. Their goal is to set `self.eval_note` and call `await self.evaluation_completed` that will then be made available to the Strategy(ies). They should be dedicated to a single simple task such as (for example) evaluate the RSI on the current data or looks for a divergence in a trend. ### Strategies Strategies are more complex elements, they can read all the evaluators evaluations on every time frame and are considering these evaluations to set their `self.eval_note` and call `await self.strategy_completed`. As a comparison if evaluators are human senses, strategies are the brain that will take these senses' signals and decide to do something or not. Strategies can be generic like SimpleStrategyEvaluator that will take any standard evaluator and time frame into account or using specific evaluators only like MoveSignalsStrategyEvaluator. ### Trading modes [Trading modes](../octobot-trading-modes/trading-modes) use the strategy(ies) evaluations to create, update or cancel orders. Using the strategies signals, they are responsible for the way to translate a signal into an order by looking at the available funds, open orders, considering stop loss or not and other trading related responsibilities. ### Triggers Evaluators, strategies and trading modes are automatically triggered when their channel has a new data. Trigger sources are: For evaluators For strategies - After a technical evaluator cycle: when all TA have updated their evaluation and called `await self.evaluation_completed` - After any real time evaluator evaluation and call of `await self.evaluation_completed` - After any social evaluator evaluation and call of `await self.evaluation_completed` For trading mode - After any strategy evaluation and call of `await self.strategy_completed` _Thanks for reading this guide and if you have any idea on how to improve it, please reach out to us !_
Shown in full with attribution under the source's licence. Licence: GPL-3.0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.