Depracated - Software for gathering historical data from Amazon Mechanical Turk Service
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
app
bin
deployment
doc
static
templates
.gitignore
.project
LICENSE.txt
README.rst

README.rst

This is depracated software. For the most recent version of the MTurk Tracker see https://github.com/ipeirotis/mturk-tracker-gae (note: that version also stopped working on February 29th, 2016)

Mturk-Tracker

Mturk-Tracker is a Django web application and it is setup within python virtualenv and requires a working python environment, PostgreSQL database and Solr search engine.

Source / How to get it

Visit github repository at

https://github.com/ipeirotis/Mturk-Tracker/

What's new

Search Crawler improvements

What's included?

The application consists of two main components:

Documentation

This file is only an introduction to the project and is also avalable at:

http://mturk-tracker.com/documentation/

whereas the complete project documentation can be found at:

http://mturk-tracker.com/documentation/mturk-tracker/

The web api documentation is available at:

http://mturk-tracker.com/documentation/mturk-tracker/api/

How does it work?

Website

As it mentioned before Mturk-Tracker site is written in Django Python web framework.

Crawlers

Amazon mturk site mturk.com is searched for available hit groups on a regular basis (approx each 6 minutes). For each encountered hit group, it's current status (number of hits available) and, if possible, if content are downloaded and saved to the database.

The data is further processed, preparing it for display and search.

Automatic project setup

This option is the easiset mean of installing the application on a target machine, however it will install all the components as there is currently no mean of excluding certain modules.

Currently deployment scripts should cover everything you need to setup the project. Only basic system configuration is required. First, create an user account which will have ownership to the project files and run ssh-server deamon.

Most things in the project are parametrized. The project configuration is based on json files that contain these parameters. Code responsible for deployment process exists in path/to/Mturk-Tracker/deployment/ subdirectory. There is a fabric script responsible for most of deployment logic and templates of configuration files that are filled with deployment parameters.

Two major files that contains deployment parameters:

  1. default.json - contains settings common across all project instances, such as the repository URL.
  2. site-name.json - contains settings specific for certain project. It also contains sensitive data thus it does not exist in the repository.

Data from this files is used for generation of resultant configuration files and scripts on the destination machine.

The structure of default.json file

This file exists in the repository. It contains parameters common for all project instances. Only several parameters are usually overriden by site-name.json that is described in the next section.

The structure of site-name.json file

This file should contain instance-specyfic deployment parameters:

  • projects_dir - path to the destination directory where the project files are to be stored;
  • settings_name - name of the resultant settings file, that will be created from settings template.
  • locals_path - path to the additional custom Django settings file that you want to upload on the destination machine;
  • db_name - name of the database;
  • db_user - name of the database user used by crawler;
  • db_password - password for the crawler database user;
  • solr_db_user - name of the database user used by Solr during data import;
  • sorl_db_password - password for the Solr database user;
  • pip_requirements - list files that contanins set of Python packages to be installed by pip;
  • system_requirements - files that constains lists of system packages to be installed by system package manager;
  • branch - name of the repository branch;
  • pip_cache - path to the directory where pip packages are cached on the destination machine;

Example site-name.json file:

{
        "projects_dir": "/data/projects",
        "settings_name": "site-name",
        "db_name": "db",
        "db_user": "db_user",
        "db_password": "****",
        "solr_db_user": "db_solr_user",
        "solr_db_password": "*****",
        "pip_requirements": ["base.txt", "devel.txt", "production.txt", "tests.txt"],
        "system_requirements": ["system_requirements.txt"],
        "locals_path": /path/to/your/requirements/on/local/machine/local.py",
        "branch": "master"
}

The structure of custom Django settings file

For better customization there is possibility to upload the file called local.py that can override some project parameters (for example in this way you can disable page caching). This file is uploaded during each deployment to the settings directory. It is imported to the main Django settings file.

Example local settings file may look like the following:

import os
from defaults import DATABASES, PROJECT_PATH, ROOT_PATH

MEDIA_ROOT = os.path.join(ROOT_PATH, 'media')
STATIC_ROOT = os.path.join(PROJECT_PATH, 'collected_static')
STATIC_URL = '/static/'

TIME_ZONE = 'UTC'
CACHE_BACKEND = 'dummy:///'

DB = DATABASES['default']
DATABASE_NAME = DB['NAME']
DATABASE_USER = DB['USER']
DATABASE_PASSWORD = DB['PASSWORD']

MTURK_AUTH_EMAIL = 'user@email.com'
MTURK_AUTH_PASSWORD = '******'

USE_CACHE = True

Running the deployment script

If all is already properly configured go to the deployment subdirectory:

$ cd path/to/Mturk-Tracker/deployment/

and run the fabric script. The following command will install all requirements (including Python and system packages):

$ fab deploy:conf_file="path/to/site-name.json",setup_environment=True -H 127.0.0.1 -u user

or you may want to update only project source (for example in order to apply changes from the repository):

$ fab deploy:conf_file="path/to/site-name.json",requirements=False -H 127.0.0.1 -u user

For more information:

$ fab help

A comprehensive description including webserver, database and crawler can be found in documentation's section on deployment: (todo: it is not yet there) http://mturk-tracker.com/documentation/mturk-tracker/ .

Manual setup

Required libraries

Mturk-Tracker requires a number of libraries that should be installed in the operating system. The most recent and complete list can be found in https://github.com/ipeirotis/Mturk-Tracker/tree/master/deployment/files/requirements/system_requirements.txt

To install the requirements simply type (for debian-like system):

$ sudo apt-get install postgresql-9.1 postgresql-server-dev-9.1 git
subversion mercurial python2.7-dev libevent-dev openjdk-6-jdk

Python environment

There is a number of python dependencies required for the project to run, see files in https://github.com/ipeirotis/Mturk-Tracker/tree/master/deployment/files/requirements/.

The easiest way to install and manage python packages is by using pip for installing packages and virtualenv for creating a separate virtual environment. If using virtualenv, consider virtualenvwrapper packed for convenience.

First, create and activate new virtual python environment:

$ virtualenv mturk --no-site-packages
$ cd  mturk
$ . bin/activate

or:

$ mkvirtualenv mturk --no-site-packages
$ workon mturk
$ cd $VIRTUAL_ENV  # optional

if using virtualenvwrapper.

TODO: Update the remainder of the guide.

After that, clone mturk code from repository and install all dependencies using pip (you have to install mercurial and subversion first, a mentioned in Required Libraries):

$ git clone https://github.com/ipeirotis/Mturk-Tracker.git
$ cd src
$ git fetch
$ git checkout -b virtualenv --track origin/virtualenv
$ echo "mturk.settings.base" > DJANGO_SETTINGS_MODULE
$ pip install -r requirements.txt

Libraries update

Because pip should take care of all libraries, use it to update already existing configuration. Whenever new dependency appears, run pip -r requirements.txt just to update.

Choosing custom settings module

By default mturk.settings.defaults configuration module is being used. To add custom variables you can add code to:

  • mturk.settings.default - project default variables visible for all other configuration files

You can also setup any other configuration module by setting DJANGO_SETTINGS_MODULE shell variable or file as given in example above.

Setting up Database

Make sure that django app can connect to database, the best way to do that is to allow postgres to accept local connections by editing pg_hba.conf file. Check if you can connect to database:

$ psql -U postgres

In order to setup a clean db you have to create the database and populate it with tables:

$ createdb -U postgres  mturk_tracker
$ createlang plpgsql -U postgres -d mturk_tracker
$ python manage.py syncdb
$ python manage.py migrate

Running django appliaction

Nothing special, just type:

$ sudo python manage.py runserver

in django project directory. And then point your browser to http://localhost:8000/

Crawling mturk

You may launch initial crawl by:

$ python manage.py crawl --workers=6 --logconf=logging.conf

Logs will be saved in /tmp/crawler.log. Because mturk requires authentication for HITs listings pagination, use --mturk-email and --mturk-password flags to authenticate and crawl as mturk worker.

To generate data that will be displayed on graphs you need to launch scripts:

$ python manage.py db_refresh_mviews
$ python manage.py db_update_agregates
$ python manage.py db_calculate_daily_stats

Searching in collected crawls

Mturk-Tracker gives ability for searching in mturk projects. Internally it uses Django-haystack application which in turn uses Solr (in version 3.6.0) indexing server as a backend.

Solr setup

Go to the Solr's page http://lucene.apache.org/solr/ for information on how to obtain appropirate Solr release.

For proper Solr's core configuration simply copy directory https://github.com/ipeirotis/Mturk-Tracker/tree/master/deployment/files/solr/solr/ to path/to/your/solr/ and manually replace the following

user="%(solr_db_user)s"
password="%(solr_db_password)s"

with

user="your_solr_db_user"
password="your_solr_db_password"

in file path/to/your/solr/solr/en/conf/import_db_hits.xml (this is done automatically in the case of automatic project setup). Next restart Solr server and visit http://127.0.0.1:8983/solr/en/select?q= (an empty xml response should be returned).

Populating the search index

If the Solr server is properly configured and some crawler data is generated try to fill up the index with the following command

python manage.py solr_data_import --verbose

You can also check Solr's status at any time. Simply type

python manage.py solr_status