Skip to content

oyeb/phoenix

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

97 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Phoenix (AI Challenge backend for Anokha 2017)

saber shall be reborn as phoenix this year 😉.

Games are configured by a single JSON file and are run inside the environment of an arena. Here's a description of what is to be contained in each directory:

├──api
│  └── The APIs, one in each folder.
├──errorlogs
│  └── Unique error logs for each game (DB stores filename alone)
├──debuglogs
│  └── Unique debug logs for each game (DB stores filename alone)
├──replays
│  └── Unique replays for each game (DB stores filename alone)
├──maps
│  └── The maps
├──bots
│  └── The bots submitted by users. They implement `__main__.py`
├──sample_bots
│  └── The bots created by us. They implement `__main__.py`
└──arena
   └── temp folder(s)

Prepare your environment

  1. Create the following directories (write a shell script for this?)
mkdir -p out/errorlogs
mkdir out/debuglogs
mkdir out/movelogs
mkdir out/replays
  1. Move the Bot APIs to /api. So this is how your /api dir should look:
api
├── python2
│   └── src
│       ├── botapi.py
│       └── __init__.py
└── python3
    └── src
        ├── botapi.py
        └── __init__.py

You can see that this folder lacks __main__.py.

  1. Copy the __main__.py of the bots into /bots
bots
├── py2_bot.py
└── py3_bot.py
  1. Create atleast 1 arena folder, this is where bot codes are copied, and logs are generated. gameloop.commit moves them to the /out folders.
mkdir arena_a

Instructions for running the Game locally:

NEW

  • Add a game_config.json to the root dir. Here's an example, <PHOENIX> stands for absolute path-to-phoenix
{
    "bots": [
        ["kevin", ["/usr/bin/python2", "python2", "<PHOENIX>/bots/py2_bot.py"]],
        ["joker", ["/usr/bin/python3", "python3", "<PHOENIX>/bots/py3_bot.py"]]
    ],
    "map": "<PHOENIX>/maps/fakemap.json",
    "arena" : "<PHOENIX>/arena",
    "filenames" : {
        "error" : "elog.json",
        "debug" : "dlog.json",
        "replay" : "replay.json",
        "move" : "moves.json"
    },
    "commit_paths" : {
        "error" : "<PHOENIX>/out/errorlogs",
        "debug" : "<PHOENIX>/out/debuglogs",
        "replay" : "<PHOENIX>/out/replays",
        "move" : "<PHOENIX>/out/movelogs"
    },
    "max_iters" : 1000,
    "timeout" : 2,
    "api_dir" : "<PHOENIX>/api"
}
  • Run python2 src

OLD

  • modify bots_config.json and map_config.json appropriately
  • run the command "python src" from phoenix directory.

Dependencies

  • libseccomp (install it from source)
  • cython (pip)
  • python-dev (for Ubuntu), python-devel (for Fedora)
  • libcap-dev (for Ubuntu), libcap-devel (for Fedora)
  • python-prctl (pip)
  • redhat-rpm-config (for Fedora only)
  • build-essential (for Ubuntu), glibc-devel (for Fedora)
  • jsonschema (pip)
  • tqdm (pip)
  • This code can only run on linux and not any unix.

##TODO

  • Define API for writing the bot (lets call this Titanic). This will be game dependent and language dependent. For example move_left() or split_two().
  • Add more syscalls to syscalls_filter(). Such as networking related ones etc.
  • Fine tune the syscalls like open etc. So that they can only open some stuff.
  • Documentation with Sphinx(low priority)
  • Look at the issues
  • Create a pip package(low priority)
  • need proper exception handling

#Architecture

Game <---> Engine <-.
                    |
                    +--> Bot
                    |
                    +--> Bot
                    |
                    .
                    .

Hence we require 2 APIs

  • Game-Engine Interface (let's call it terra) is the backbone of our operation.
    • Game gets inputs, transforms its "state", and returns it to the Engine.
    • Engine can query status, scores, etc from Game.
  • Engine-Bot Interface (let's call this sky)
    • Parses Bot moves (possibly even validates them)
    • Bot gets game-state, commands and logs.

So,

  • Bot gets Game details by sky.
  • Engine uses sky to control Bot process and its function and also to handover Game state, score and logs to them.
  • Engine uses terra to send Bot moves and Bot status to Game.
  • Game uses terra to return latest game-state, scores and logs to Engine which are forwarded).

Whatever be the Game, we can start work on the APIs and the Engine. We already know what the responsibilities of the Engine are.

#Engine

Responsibilities

  • Start Game process.
    • Also open log file handles
  • Start Bot processes.
  • Pass Game State to Bots.
  • Pass moves to Game
  • Pass end-game logs, scores etc. to Bots
  • Handling Bot errors
    • Unresponsive, Invalid or even Malicious
  • Security
  • Subprocess management
    • start, end, resource (mem, cpu, "real" time), privileges

Bots

  • The bots' information must be stored in bots_config.json as JSON objects. They must be directly usable by system calls such as exec*

#Security

We can protect our system by running bot process in chroot jails or containers (LXD, LXC or Docker) which help in minimising damages when scripts attempt privilege escalation or arbit code injection.

How do we make sure processes cannot open any kind of file descriptors? Be it for IPC or network sockets?

About

The AI Challenge backend for Anokha 2017

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages