Project Board | Community Site | #openverse @ Slack | Handbook | Figma Mockups | Figma Design Library
This repository contains the methods used to identify over 1.4 billion Creative Commons licensed works. The challenge is that these works are dispersed throughout the web and identifying them requires a combination of techniques.
Two approaches are currently in use:
- Web crawl data
- Application Programming Interfaces (API Data)
The Common Crawl Foundation provides an open repository of petabyte-scale web crawl data. A new dataset is published at the end of each month comprising over 200 TiB of uncompressed data.
The data is available in three file formats:
- WARC (Web ARChive): the entire raw data, including HTTP response metadata, WARC metadata, etc.
- WET: extracted plaintext from each webpage.
- WAT: extracted html metadata, e.g. HTTP headers and hyperlinks, etc.
For more information about these formats, please see the Common Crawl documentation.
Openverse Catalog uses AWS Data Pipeline service to automatically create an Amazon EMR cluster of 100 c4.8xlarge instances that will parse the WAT archives to identify all domains that link to creativecommons.org. Due to the volume of data, Apache Spark is used to streamline the processing. The output of this methodology is a series of parquet files that contain:
- the domains and its respective content path and query string (i.e. the exact webpage that links to creativecommons.org)
- the CC referenced hyperlink (which may indicate a license),
- HTML meta data in JSON format which indicates the number of images on each webpage and other domains that they reference,
- the location of the webpage in the WARC file so that the page contents can be found.
The steps above are performed in ExtractCCLinks.py
.
Apache Airflow is used to manage the workflow for various API ETL jobs which pull and process data from a number of open APIs on the internet.
To view more information about all the available workflows (DAGs) within the project, see DAGs.md.
See each provider API script's notes in their respective handbook entry.
There are a number of scripts in the directory
openverse_catalog/dags/provider_api_scripts
eventually
loaded into a database to be indexed for searching in the Openverse API. These run in a
different environment than the PySpark portion of the project, and so have their
own dependency requirements.
For instructions geared specifically towards production deployments, see DEPLOYMENT.md
You'll need docker
and docker-compose
installed on your machine, with
versions new enough to use version 3
of Docker Compose .yml
files.
You will also need the just
command runner installed.
To set up the local python environment along with the pre-commit hook, run:
python3 -m venv venv
source venv/bin/activate
just install
The containers will be built when starting the stack up for the first time. If you'd like to build them prior to that, run:
just build
To set up environment variables run:
just dotenv
This will generate a .env
file which is used by the containers.
The .env
file is split into four sections:
- Airflow Settings - these can be used to tweak various Airflow properties
- API Keys - set these if you intend to test one of the provider APIs referenced
- Connection/Variable info - this will not likely need to be modified for local development, though the values will need to be changed in production
- Other config - misc. configuration settings, some of which are useful for local dev
The .env
file does not need to be modified if you only want to run the tests.
There is a docker-compose.yml
provided in the
openverse_catalog
directory, so from that directory, run
just up
This results, among other things, in the following running containers:
openverse_catalog_webserver_1
openverse_catalog_postgres_1
openverse_catalog_s3_1
and some networking setup so that they can communicate. Note:
openverse_catalog_webserver_1
is running the Apache Airflow daemon, and also has a few development tools (e.g.,pytest
) installed.openverse_catalog_postgres_1
is running PostgreSQL, and is setup with some databases and tables to emulate the production environment. It also provides a database for Airflow to store its running state.- The directory containing all modules files (including DAGs, dependencies, and other
tooling) will be mounted to the directory
/usr/local/airflow/openverse_catalog
in the containeropenverse_catalog_webserver_1
. On production, only the DAGs folder will be mounted, e.g./usr/local/airflow/openverse_catalog/dags
.
The various services can be accessed using these links:
- Airflow:
localhost:9090
(The default username and password are bothairflow
.) - Minio Console:
localhost:5011
(The default username and password aretest_key
andtest_secret
) - Postgres:
localhost:5434
(using a database connector)
At this stage, you can run the tests via:
just test
# Alternatively, run all tests including longer-running ones
just test --extended
Edits to the source files or tests can be made on your local machine, then tests can be run in the container via the above command to see the effects.
If you'd like, it's possible to login to the webserver container via:
just shell
If you just need to run an airflow command, you can use the airflow
recipe. Arguments passed to airflow must be quoted:
just airflow "config list"
To follow the logs of the running container:
just logs
To begin an interactive pgcli
shell on the database container, run:
just db-shell
If you'd like to bring down the containers, run
just down
To reset the test DB (wiping out all databases, schemata, and tables), run
just down -v
docker volume prune
can also be useful if you've already stopped the running containers, but be warned that it will remove all volumes associated with stopped containers, not just openverse-catalog ones.
To fully recreate everything from the ground up, you can use:
just recreate
openverse-catalog
├── .github/ # Templates for GitHub
├── archive/ # Files related to the previous CommonCrawl parsing implementation
├── docker/ # Dockerfiles and supporting files
│ ├── airflow/ # - Docker image for Airflow server and workers
│ └── local_postgres/ # - Docker image for development Postgres database
├── openverse_catalog/ # Primary code directory
│ ├── dags/ # DAGs & DAG support code
│ │ ├── common/ # - Shared modules used across DAGs
│ │ ├── commoncrawl/ # - DAGs & scripts for commoncrawl parsing
│ │ ├── database/ # - DAGs related to database actions (matview refresh, cleaning, etc.)
│ │ ├── maintenance/ # - DAGs related to airflow/infrastructure maintenance
│ │ ├── oauth2/ # - DAGs & code for Oauth2 key management
│ │ ├── providers/ # - DAGs & code for provider ingestion
│ │ │ ├── provider_api_scripts/ # - API access code specific to providers
│ │ │ └── *.py # - DAG definition files for providers
│ │ └── retired/ # - DAGs & code that is no longer needed but might be a useful guide for the future
│ └── templates/ # Templates for generating new provider code
└── * # Documentation, configuration files, and project requirements
The docker image for the catalog (Airflow) is published to ghcr.io/WordPress/openverse-catalog.
Pull requests are welcome! Feel free to join us on Slack and discuss the project with the engineers and community members on #openverse.
Openverse, previously known as CC Search, was conceived and built at Creative Commons. We thank them for their commitment to open source and openly licensed content, with particular thanks to previous team members @ryanmerkley, @janetpkr, @lizadaly, @sebworks, @pa-w, @kgodey, @annatuma, @mathemancer, @aldenstpage, @brenoferreira, and @sclachar, along with their community of volunteers.