Garzón is a "judge" system, an automatic problem corrector. You submit solutions to problems and Garzón accepts or rejects them. It is inspired by the way programming contests work. The system has three parts: a load balancer (the 'server'), several workers, and a virtual machine or VM that each worker controls:

There are separate directories for each part:

• grz-vm: a script to manage VMs.
• grz-worker: a binary that connects to the server and controls the VMs (a QEmu process).
• server: a small library for the server process.

This script assists in creating virtual machines. To reduce the footprint of the VM, Tiny Core Linux is used. This allows the creation of very small and quick virtual machines. The typical steps in creating a virtual machine are:

$ grz-vm download
$ grz-vm remaster core.gz initrd.gz
$ grz-vm createimg garzon.img 400
$ grz-vm install garzon.img go gcc
$ grz-vm convert garzon.img garzon.qcow2

The first step downloads the kernel and initrd image for Tiny Core Linux. The "remaster" part introduces modifications to the initrd specific to Garzón. Then a new disk image is created (garzon.img), some software is installed within, and finally the disk image is converted to the qcow format, suitable for snapshotting with QEmu.

The worker first launches a non-graphical instance of QEmu, which will look for the environment variable GARZON_VMS, and will by default use garzon.qcow2 as image. This enables the use of different images for different purposes. The options are:

$ grz-worker -help
Usage of grz-worker:
  -graphic=false: Show QEmu graphic mode
  -image="garzon.qcow2": Specify image file to use
  -prepare=false: Only create the snapshot

You can 1) see the QEmu console using -graphic=true, 2) specify the image, and 3) for the first run, do a -prepare, which does a snapshot of the clean state of the virtual machine.

The server library has a very simple API. It only needs one call to:

server.Handle()

during initialization. This will register an HTTP handler to accept connections from workers.

When evaluating a submission, you just call:

func Judge(submission Submission, report func(msg string)) (veredict string, err error)

This function will evaluate the submission using one of the available workers, optionally notify progress by using the report callback, and will return the veredict (or an error).

There is an example server which implements a very simple web application using websockets that lists problems and lets users register and send problems (no DB).