jupyter_grade_server

An EdX auto grading server for Jupyter notebooks based on xqueue_watcher, an implementation of a polling XQueue client and grader.

Main project: edx_notebook_grader

Running

python -m jupyter_grade_server -d [path to settings directory]

JSON configuration file

{
	"test-123": {
		"SERVER": "http://127.0.0.1:18040",
		"CONNECTIONS": 1,
		"AUTH": ["lms", "lms"],
		"HANDLERS": [
			{
				"HANDLER": "xqueue_watcher.grader.Grader",
				"KWARGS": {
					"grader_root": "/path/to/course/graders/",
				}
			}
		]
	}
}

• test-123: the name of the queue
• SERVER: XQueue server address
• AUTH: list of username, password
• CONNECTIONS: how many threads to spawn to watch the queue
• HANDLERS: list of callables that will be called for each queue submission
  • HANDLER: callable name
  • KWARGS: optional keyword arguments to apply during instantiation

xqueue_watcher.grader.Grader

To implement a pull grader:

Subclass xqueue_watcher.grader.Grader and override the grade method. Then add your grader to the config like "handler": "my_module.MyGrader". The arguments for the grade method are: * grader_path: absolute path to the grader defined for the current problem * grader_config: other configuration particular to the problem * student_response: student-supplied code

Sandboxing

To sandbox python, use CodeJail. In your handler configuration, add:

"CODEJAIL": {
	"name": "python",
	"python_bin": "/path/to/sandbox/python",
	"user": "sandbox_username"
}

Then, codejail_python will automatically be added to the kwargs for your handler. You can then import codejail.jail_code and run jail_code("python", code...). You can define multiple sandboxes and use them as in jail_code("special-python", ...)