skoocda / DistributedReplays

👨‍💻 Replay Collection and Analysis Server for Saltie DRL Bot

Saltie Replay Server

A replay collection server for Saltie the Deep Reinforcement Learning Bot

Modeled after the Leela Zero distributed Go engine.

Setup

• Install Flask (flask).
• Install Flask login (flask_login)
• Install psycopg2 (psycopg2)
• Install postgreSQL
• If only for local development change the password to postgres
• Run RLBotServer.py.

The server currently only allows uploads from a single IP address every 4.5 minutes, to prevent abuse. This is also in-line with the data generation speed of a typical Rocket League game (~5 minutes). Each replay has an IP address attached to it, so blame can be given.

##Windows

• Install python/pip
• Install postgreSQL
• Install RabbitMQ
• Run win_setup.bat
• Run win_run.bat

Replays

Replays will be saved in /replays for later consumption. The server provides these files to clients.

Structure

The structure of the server is split into different files:

• celery_tasks.py - this contains the tasks that run in the background (i.e. the replay parsing / pickling) using Celery workers
• celeryconfig.py - configuration for the Celery worker
• constants.py - things like dict of id -> car body name
• functions.py - general helper functions for the server
• middleware.py - classes that are used in the actual inner framework code of the server
• objects.py - SQL ORM objects such as the Replay and Model objects
• replays.py - contains all code pertaining to replay parsing (besides the Celery task)
• saltie.py - contains all code pertaining to the Saltie data collection part
• startup.py - is run to start the connection to the SQL server
• stats.py - just some analysis on the upload patterns of the users