ghtorrent: Mirror and index data from the Github API
A library and a collection of scripts used to retrieve data from the Github API
and extract metadata in an SQL database, in a modular and scalable manner. The
scripts are distributed as a Gem (ghtorrent), but they can also be run by
checking out this repository.
GHTorrent can be used for a variety of purposes, such as:
- Mirror the Github API event stream and follow links from events to actual data
to gradually build a Github index - Create a queriable metadata database for a specific repository
- Construct a data source for extracting process analytics (see for example those) for one or more repositories
Components
GHTorrents components (which can be used individually) are:
- APIClient: Knows how to query the Github API (both single entities and
pages) and respect the API request limit. Can be configured to override the
default IP address, in case of multihomed hosts. - Retriever: Knows how to retrieve specific Github entities (users, repositories, watchers) by name. Uses an optional persister to avoid
retrieving data that have not changed. - Persister: A key/value store, which can be backed by a real key/value store,
to store Github JSON replies and query them on request. The backing key/value
store must support arbitrary queries to the stored JSON objects. - GHTorrent: Knows how to extract information from the data retrieved by
the retriever in order to update an SQL database (see schema) with metadata.
Component Configuration
The Persister and GHTorrent components have configurable back ends:
- Persister: Either uses MongoDB > 3.0 (
mongodriver) or no persister (noopdriver) - GHTorrent: GHTorrent is tested mainly with MySQL and SQLite, but can theoretically be used with any SQL database compatible with Sequel. Your milaege may vary.
For distributed mirroring you also need RabbitMQ >= 3.3
Installation
1. Install GHTorrent
GHTorrent is written in Ruby (tested with Ruby > 2.0). To install it as a Gem do:
2. Install Your Preferred Database
Depending on which SQL database you want to use, install the appropriate
dependency gem.
Configuration
Copy config.yaml.tmpl
to a file in your home directory.
All provided scripts accept the -c option, which accepts the location of the configuration file as
a parameter.
You can find more information of how you can setup a mirroring cluster of machines
to retrieve data in parallel on the Wiki.
Using GHTorrent
To mirror the event stream and capture all data:
-
ght-mirror-events.rbperiodically polls Github's event
queue (https://api.github.com/events), stores all new events in the
configured pestister, and posts them to thegithubexchange in
RabbitMQ. -
ght-data_retrieval.rbcreates queues that route posted events to processor
functions. The functions use the appropriate Github API call to retrieve the
linked contents, extract metadata (for database storage), and store the
retrieved data in the appropriate collection in the persister, to avoid
duplicate API calls.
Data in the SQL database contain pointers (theext_ref_idfield) to the
"raw" data in the persister.
To retrieve data for a repository or user:
ght-retrieve-reporetrieves all data for a specific repositoryght-retrieve-userretrieves all data for a specific user
To perform maintenance:
ght-loadloads selected events from the persister to the queue in order for
theght-data-retrievalscript to reprocess them
Data
The code in this repository is used to power the data collection process of
the GHTorrent.org project.
You can find all data collected by in the project in the
Downloads page.
There are two sets of data:
- Raw events: Github's event stream. These
are the roots for mirroring operations. Theght-data-retrievalcrawler starts
from an event and goes deep into the rabbit hole. - SQL dumps + Linked data: Data dumps from the SQL database and the corresponding MongoDB entities.
Bugs & Feature Requests
Please tell us about features you'd like or bugs you've discovered on our
Issue Tracker.
Patches, bug fixes, etc are welcome. Please fork the repository and create
a pull request when done fixing/implementing the new feature.
Citing GHTorrent in your Research
If you find GHTorrent and the accompanying datasets useful in your research,
please consider citing the following paper:
Georgios Gousios and Diomidis Spinellis, "GHTorrent: GitHub’s data from a firehose," in MSR '12: Proceedings of the 9th Working Conference on Mining Software Repositories, June 2-–3, 2012. Zurich, Switzerland.