recodex-wiki/Fileserver.md

Fileserver

The fileserver is a simple frontend to a disk storage space that contains auxiliary files for assignments, archives with job configuration and files submitted by users and evaluation results. These files are the only ones required for backend to run, so dedicated fileserver gives the possibility of testing backend separately. Also, one fileserver instance could be shared among multiple API instances (with the same broker), so common files does not need to be duplicated in each API instance.

One exception is that important files with character of database entry (but not stored in database due to size) are stored directly in filesystem of API server. But this fact does not devaluate benefit of separate fileserver. From security point of view, fileserver should be completely isolated from public internet to keep the data safe while API server must be public from its nature.

For a description of the communication protocol used by the frontend and workers, see the Communication chapter.

Description

The storage is implemented in Python, using the Flask web framework. This particular implementation evolved from a simple mock fileserver we used in early stages of development. It prooved to be very reliable, so we decided to keep fileserver as separate component instead of integrating this functionality into main API.

Internal storage structure

Fileserver stores its data in a configurable filesystem folder. This folder has the following subfolders:

• ./submissions/<id> -- folders that contain files submitted by users (student's solutions to assignments). <id> is an identifier received from the ReCodEx API.
• ./submission_archives/<id>.zip -- ZIP archives of all submissions. These are created automatically when a submission is uploaded. <id> is an identifier of the corresponding submission.
• ./tasks/<subkey>/<key> -- supplementary task files (e.g. test inputs and outputs). <key> is a hash of the file content (sha-1 is used) and <subkey> is its first letter (this is an attempt to prevent creating a flat directory structure).

Installation

To install and use the fileserver, it is necessary to have Python3 with pip package manager installed. It is needed to install the dependencies. From clonned repository run the following command:

$ pip install -r requirements.txt

That is it. Fileserver does not need any special installation. It is possible to build and install rpm package or install it without packaging the same way as monitor, but it is only optional. The installation would provide you with script recodex-fileserver in you PATH. No systemd unit files are provided, because of the configuration and usage of fileserver component is much different to our other Python parts.

Configuration and usage

There are several ways of running the ReCodEx fileserver. We will cover two typical use cases.

Running in development mode

For simple development usage, it is possible to run the fileserver in the command line. Allowed options are described below.

usage: fileserver.py [--directory WORKING_DIRECTORY]
                     {runserver,shell} ...

• runserver argument starts the Flask development server (i.e. app.run()). As additional argument can be given a port number.
• shell argument instructs Flask to run a Python shell inside application context.

Simple development server on port 9999 can be run as

$ python3 fileserver.py runserver 9999

When run like this command, the fileserver creates a temporary directory where it stores all the files and which is deleted when it exits.

Running as WSGI script in a web server

If you need features such as HTTP authentication (recommended) or efficient serving of static files, it is recommended to run the app in a full-fledged web server (such as Apache or Nginx) using WSGI. Apache configuration can be generated by mkconfig.py script from the repository.

usage: mkconfig.py apache [-h] [--port PORT] --working-directory
                          WORKING_DIRECTORY [--htpasswd HTPASSWD]
                          [--user USER]

• port -- port where the fileserver should listen
• working_directory -- directory where the files should be stored
• htpasswd -- path to user file for HTTP Basic Authentication
• user -- user under which the server should be run

Running using uWSGI

Another option is to run fileserver as a standalone app via uWSGI service. Setup is also quite simple, configuration file can be also generated by mkconfig.py script.

1. (Optional) Create a user for running the fileserver

2. Make sure that your user can access your clone of the repository

3. Run mkconfig.py script.

   usage: mkconfig.py uwsgi [-h] [--user USER] [--port PORT]
                            [--socket SOCKET]
                            --working-directory WORKING_DIRECTORY
   

   • user -- user under which the server should be run
   • port -- port where the fileserver should listen
   • socket -- path to UNIX socket where the fileserver should listen
   • working_directory -- directory where the files should be stored

4. Save the configuration file generated by the script and run it with uWSGI, either directly or using systemd. This depends heavily on your distribution.

5. To integrate this with another web server, see the uWSGI documentation

Note that the ways distributions package uWSGI can vary wildly. In Debian 8 it is necessary to convert the configuration file to XML and make sure that the python3 plugin is loaded instead of python. This plugin also uses Python 3.4, even though the rest of the system uses Python 3.5 - make sure to install dependencies for the correct version.