Databases provide a critical aspect of psiTurk, as they store data from experiments and help to prevent the same user from completing your experiment more than once. Databases provide an important function for web-based experiments – Because multiple people can complete your experiment at the same time, you need a system which can simultaneously write/read data. Databases are optimized for this type of environment and are thus very useful for experiments.
psiTurk can integrate with any database that is compatible with SQLAlchemy.
database_url – For details on how to configure databases in config.txt
Perhaps the simplest quickstart solution is to use SQLite. This database solution writes to a local file on the same computer as is running the psiTurk server.
To use a SQLite data base, simply set the database_url field in your local configuration file (config.txt):
database_url = sqlite:///FILENAME.db
where FILENAME is of your choosing. By default, psiTurk sets this like this:
database_url = sqlite:///participants.db
This will make a SQLite database file in the top-level folder of your project. If you change the database_url and restart psiTurk, a new database corresponding to the new filename will be created. If you set it to an existing file name, psiTurk will attempt to connect to this database.
It is best to do this while the server is not running (note in this example the “server” status says “off”). If you change this while the server is running you will need to type:
[psiTurk server:on mode:sdbx #HITs:0]$ server restart
While great for development and debugging, SQLite has a number of important downsides for deploying experiments. In particular, SQLite does not allow concurrent access to the database, so if the locks work properly, simultaneous access (say, from multiple users submitting their data at the same time) could destabilize your database. In the worst scenario, the database could become corrupted, resulting in data loss.
As a result, we recommend using a more robust database solution when actually running your experiment.
However, SQLite is a good solution particularly for initial testing. It is also possible to try to “throttle” the rate of signups on Mechanical Turk (by only posting one assignment slot at a time) so that database errors are less likely using SQLite.
SQLite database are fine for local testing but more robust databases like MySQL are recommended especially if you plan to run many participants simultaneously. Again, any server compatible with SQLAlchemy can be used.
The Heroku free tier includes access to a postgresql database. See Running psiTurk on Heroku.
Similar to postgresql on Heroku, another robust solution is to set up a MySQL database. psiTurk’s reliance on SQLAlchemy for interfacing with database which means it is easy to switch between MySQL, PostgreSQL, or SQLite.
For example, to use an existing MySQL database:
database_url = mysql://USERNAME:PASSWORD@HOSTNAME:PORT/DATABASE
where USERNAME and PASSWORD are your access credentials for the database, HOSTNAME is the DNS entry or IP address for the database, PORT is the port number (default is 3306) and DATABASE is the name of the database on the server.
Use 127.0.0.1 as the HOSTNAME for a database running locally to the psiTurk server rather than ‘localhost’. Mysql treats the HOSTNAME ‘localhost’ as a special case in Unix-based systems and will cause the psiTurk server to fail to boot.
Here’s an example of setting up a minimal MySQL database for use with psiTurk:
$ mysql -uroot -p
mysql> CREATE USER 'your_username'@'localhost' IDENTIFIED BY 'your_password';
Query OK, 0 rows affected (0.03 sec)
mysql> CREATE DATABASE your_database;
Query OK, 1 row affected (0.01 sec)
mysql> GRANT ALL PRIVILEGES ON your_database.* TO 'your_username'@'localhost';
Query OK, 0 rows affected (0.00 sec)
where your_username, your_password and your_database match the USERNAME, PASSWORD and DATABASE specified in config.txt’s database_url variable.
The table specified in config.txt:
assignment_table_name = turkdemo
…will be created automatically when running the psiturk shell. In addition several other tables are automatically created to manage other aspects of task scheduling and hit management (see the hits_table_name, campaigns_table_name, and jobs_table_name options in config.txt).
MySQL is (fairly) easy to install and free. However, a variety of web hosting services offer managed MySQL databases. Some are even free.