How to download and setup data

Metadata DB of NPM Packages


Disk Space: Currently (April 2023), 200 GB is sufficient space for downloading the metadata dataset and importing it to PostgreSQL. Running PostgreSQL on SSD storage probably provides a substantial speed up.

PostgeSQL: The database currently uses PostgreSQL 15.x, though the database dumps should be importable to newer versions as well. You can find instructions for installing PostgreSQL here. Note that if you do not have root, then building from source allows you to use PostgreSQL as non-root.

Additionally, it is recommended to tweak the default PostgreSQL configuration to achieve better performance for data analysis queries. This website (using the Data warehouse option) is a useful tool for suggesting configuration options to change.

A copy of the database: If you haven't yet, see the downloads page to download a dump of the metadata database.

Importing the NPM Package Metadata Database

Once you have the database tarball downloaded and PostgreSQL setup, first untar the file:

tar -xvf latest.tar
rm latest.tar

You should now have a directory called db_export. You can now import this to a new Postgres databased named npm_data by running:

psql postgres -c "CREATE DATABASE npm_data WITH TEMPLATE = template0;"
psql npm_data -c "DROP SCHEMA public;"
pg_restore -d npm_data -e -O -j 4 db_export/

Note that depending on your Postgres installation, you may need to run the above command as the Postgres root user, commonly postgres. Additionally, you may need to pass various connection options (port, etc.) to psql and pg_restore depending on how Postgres is configured.

The pg_restore command may take up to around an hour or more, depending on hardware.

Once pg_restore completes, you can then connect to the database using psql npm_data, or any other SQL tool. To test that the import was successful, try to count the number of packages. It should look something like this:

$ psql npm_data
psql (15.2 (Homebrew))
Type "help" for help.

npm_data=# select count(*) from packages;
(1 row)

If this all works, then you are ready to start developing your own NPM package metadata queries. See the documentation to understand the structure of the tables in the database.

Source Code of NPM Packages


Disk Space: Currently (April 2023), over 20 TB of disk space is needed to download and use the source code dataset. Please ensure that you have sufficient space first.

Metadata Database: In order to make meaningful use of the source code dataset, you must also have the metadata database setup.

Rust: You may install Rust via rustup. Version 1.66.0 or newer should work.

Python: Our code has been tested with Python 3.8.1, other versions may work as well.

Redis: Download Redis using your favorite package manager. For instance, if using snap: sudo snap install redis.

Part 1: Downloading the Blob Store

  1. Clone the npm-follower repository, and change to the directory with dataset downloading scripts:
git clone
cd npm-follower/database_exporting
  1. Install Python requirements:
# Could be pip3 depending on your system.
# If you have conflicting installations, or if you prefer, feel free to use a Pip virtual environment.
pip install -r download_tarballs_requirements.txt
  1. Download the contents of the HuggingFace repo:
# Could be python3 depending on your system.

This will proceed to download the entire 20+ TB over the network, and after that verify the contents of all files, and redownload any incorrect files. You must run the above command multiple times until it outputs Re-downloading 0 files, otherwise you will likely receive some corrupted files. Needless to say, this will take quite a while.

  1. The dataset is split into pieces so that each file is under 50 GB. You now must restore the dataset:

You should now have 1000 blob files in the /PATH/TO/DATASET/DIRECTORY directory.

Part 2: Setting up Redis

  1. Locate your Redis data directory: If you are unsure where you Redis data directory is, you may check with redis-cli config get dir. To do so you may need to start Redis, e.g.: sudo systemctl start redis-server.service.

  2. Stop Redis, e.g.: sudo systemctl stop redis-server.service.

  3. Unpack the Redis dump: the downloaded dump of the metadata database (downloaded in the previous section) includes a dump of a Redis database as well, named db_export/ You now need to unpack it to your Redis data directory:

# ***NOTE***: If different, use the path to your 
# Redis data directory in place of /var/lib/redis

sudo su
mv db_export/ /var/lib/redis
cd /var/lib/redis
chown -R redis:redis .
  1. Modify the Redis configuration file (typically /etc/redis/redis.conf) as follows:
appendonly yes
appendfilename "appendonly.aof"
appenddirname "appendonlydir"
  1. Start Redis: To start Redis and enable it to automatically run at startup, run:
sudo systemctl enable --now redis-server.service

The Redis server with the blob metastore should be running at on the Redis database number 4, at the URL: redis://

Part 3: Setting up Blob Store Server

The blob store server implements an HTTP API for retrieving and managing file metadata in the blob filesystem.

  1. Change directory to where you cloned the npm-follower repo: cd /path/to/npm-follower/
  2. Modify .env with the following settings:
  1. Modify .secret.env (create it if it doesn't exist) with the following setting:
  1. Run the Blob Store Server:
cargo run --release --bin blob_idx_server -- 0 0

The server has capabilities for running distributed jobs across multiple machines in a slurm cluster. The first two arguments correspond to the number of file-transfer workers and compute workers. In this case, we are running the server on a single machine in order to retrieve data from the blob store, so we set both of these arguments to 0.

Everything is now fully setup! Please see the documentation to learn how to read data from the blob store.