Getting started

We’re pleased that you are interested in working on Warehouse.

Your first pull request

After you set up your development environment and ensure you can run the tests and build the documentation (using the instructions in this document), take a look at our guide to the Warehouse codebase. Then, look at our open issues that are labelled “good first issue”, find one you want to work on, comment on it to say you’re working on it, then submit a pull request. Use our Submitting patches documentation to help.

Setting up a development environment to work on Warehouse should be a straightforward process. If you have any difficulty, contact us so we can improve the process:

  • For bug reports or general problems, file an issue on GitHub;

  • For real-time chat with other PyPA developers, join #pypa-dev on Libera, or the PyPA Discord;

  • For longer-form questions or discussion, visit Discourse.

Detailed installation instructions

Getting the Warehouse source code

Fork the repository on GitHub and clone it to your local machine:

git clone

Add a remote and regularly sync to make sure you stay up-to-date with our repository:

git remote add upstream
git checkout main
git fetch upstream
git merge upstream/main

Configure the development environment


In case you are used to using a venv/virtualenv or virtual environment for Python development: don’t use one for warehouse development. Our Makefile scripts and Docker container development flow creates and removes virtualenvs as needed while you are building and testing your work locally.

Why Docker?

Docker simplifies development environment set up.

Warehouse uses Docker and Docker Compose to automate setting up a “batteries included” development environment. The Dockerfile and docker-compose.yml files include all the required steps for installing and configuring all the required external services of the development environment.

Installing Docker

The best experience for building Warehouse on Windows 10 is to use the Windows Subsystem for Linux (WSL) in combination with both Docker for Windows and Docker for Linux. Follow the instructions for both platforms, and see Docker and Windows Subsystem for Linux Quirks for extra configuration instructions.

Verifying Docker installation

Check that Docker is installed: docker -v

Install Docker Compose

Install Docker Compose using the Docker-provided installation instructions.


Docker Compose will be installed by Docker for Mac and Docker for Windows automatically.

Verifying Docker Compose installation

Check that Docker Compose is installed: docker-compose -v

Verifying the necessary ports are available

Warehouse needs access to a few local ports in order to run, namely ports 80, 5433, and 9000. You should check each of these for availability with the lsof command.

For example, checking port 80:

sudo lsof -i:80 | grep LISTEN

If the port is in use, the command will produce output, and you will need to determine what is occupying the port and shut down the corresponding service. Otherwise, the port is available for Warehouse to use, and you can continue.

Building the Warehouse Container

Once you have Docker and Docker Compose installed, run:

make build

in the repository root directory.

This will pull down all of the required docker containers, build Warehouse and run all of the needed services. The Warehouse repository will be mounted inside the Docker container at /opt/warehouse/src/. After the initial build, you should not have to run this command again.

Running the Warehouse container and services

You have to start the Docker services that make up the Warehouse application.


These services need ~4 GB of RAM dedicated to Docker to work. This is more than the default setting of the Docker Engine of 2 GB. Thus, you need to increase the memory allocated to Docker in Docker Preferences (on Mac) or Docker Settings (on Windows) by moving the slider to 4 GB in the GUI.

If you are using Linux, you may need to configure the maximum map count to get the elasticsearch up and running. According to the documentation this can be set temporarily:

# sysctl -w vm.max_map_count=262144

or permanently by modifying the vm.max_map_count setting in your /etc/sysctl.conf.

Also check that you have more than 5% disk space free, otherwise elasticsearch will become read only. See flood_stage in the elasticsearch disk allocation docs.

Once make build has finished, run the command:

make serve

This command starts the containers that run Warehouse on your local machine. After the initial build process, you will only need this command each time you want to startup Warehouse locally.

make serve will produce output for a while, and will not exit. Eventually the output will cease, and you will see a log message indicating that either the web service has started listening:

web_1 | [2018-05-01 20:28:14 +0000] [6] [INFO] Starting gunicorn 19.7.1
web_1 | [2018-05-01 20:28:14 +0000] [6] [INFO] Listening at: (6)
web_1 | [2018-05-01 20:28:14 +0000] [6] [INFO] Using worker: sync
web_1 | [2018-05-01 20:28:14 +0000] [15] [INFO] Booting worker with pid: 15

or that the static container has finished compiling the static assets:

static_1 | [20:28:37] Starting 'dist:compress'...
static_1 | [20:28:37] Finished 'dist:compress' after 14 μs
static_1 | [20:28:37] Finished 'dist' after 43 s
static_1 | [20:28:37] Starting 'watch'...
static_1 | [20:28:37] Finished 'watch' after 11 ms

or maybe something else.

After the docker containers are setup in the previous step, in a separate terminal session, run:

make initdb

This command will:

  • create a new Postgres database,

  • install example data to the Postgres database,

  • run migrations,

  • load some example data from Test PyPI, and

  • index all the data for the search database.


If you get an error about xz, you may need to install the xz utility. This is highly likely on macOS and Windows.

Once the make initdb command has finished, you are ready to continue.

Viewing Warehouse in a browser

At this point all the services are up, and web container is listening on port 80. It’s accessible at http://localhost:80/.


If you are using docker-machine on an older version of macOS or Windows, the warehouse application might be accessible at https://<docker-ip>:80/ instead. You can get information about the docker container with docker-machine env


On Firefox, the logos might show up as black rectangles due to the Content Security Policy used and an implementation bug in Firefox (see this bug report for more info).

Logging in to Warehouse

In the development environment, the password for every account has been set to the string password. You can log in as any account at http://localhost:80/account/login/.

To log in as an admin user, log in as ewdurbin with the password password at http://localhost:80/admin/login/.

Stopping Warehouse and other services

In the terminal where make serve is running, you can use Control-C to gracefully stop all Docker containers, and thus the one running the Warehouse application.

Or, from another terminal, use make stop in the Warehouse repository root; that’ll stop all the Docker processes with warehouse in the name.

What did we just do and what is happening behind the scenes?

The repository is exposed inside of the web container at /opt/warehouse/src/ and Warehouse will automatically reload when it detects any changes made to the code.

The example data located in dev/example.sql.xz is taken from Test PyPI and has been sanitized to remove anything private.

Running your developer environment after initial setup

You won’t have to initialize the database after the first time you do so, and you will rarely have to re-run make build. Ordinarily, to access your developer environment, you’ll:

make serve

View Warehouse in the browser at http://localhost:80/.

Debugging the webserver

If you would like to use a debugger like pdb that allows you to drop into a shell, you can use make debug instead of make serve.


Errors when executing make build

  • If you are using Ubuntu and invalid reference format error is displayed, you can fix it by installing Docker through Snap <>.

snap install docker
  • If you receive the error: python3.8: command not found , ensure you have Python 3.8 installed on your system. This is the “base” Python version that Warehouse uses to create the rest of the development environment.

Errors when executing make serve

  • If the Dockerfile is edited or new dependencies are added (either by you or a prior pull request), a new container will need to built. A new container can be built by running make build. This should be done before running make serve again.

  • If make serve hangs after a new build, you should stop any running containers and repeat make serve.

  • To run Warehouse behind a proxy set the appropriate proxy settings in the Dockerfile.

  • If sqlalchemy.exec.OperationalError is displayed in localhost after make serve has been executed, shut down the Docker containers. When the containers have shut down, run make serve in one terminal window while running make initdb in a separate terminal window.

Errors when executing make purge

  • If make purge fails with a permission error, check ownership and permissions on warehouse/static. docker-compose is spawning containers with docker. Generally on Linux that process is running as root. So when it writes files back to the file system as the static container does those are owned by root. So your docker daemon would be running as root, so your user doesn’t have permission to remove the files written by the containers. sudo make purge will work.

Errors when executing make initdb

  • If make initdb fails with a timeout like:

    urllib3.exceptions.ConnectTimeoutError: (<urllib3.connection.HTTPConnection object at 0x8beca733c3c8>, 'Connection to elasticsearch timed out. (connect timeout=30)')

    you might need to increase the amount of memory allocated to docker, since elasticsearch wants a lot of memory (Dustin gives warehouse ~4GB locally). Refer to the tip under Running the Warehouse container and services section for more details.

“no space left on device” when using docker-compose

docker-compose may leave orphaned volumes during teardown. If you run into the message “no space left on device”, try running the following command (assuming Docker >= 1.9):

docker volume rm $(docker volume ls -qf dangling=true)


This will delete orphaned volumes as well as directories that are not volumes in /var/lib/docker/volumes

(Solution found and further details available at

make initdb is slow or appears to make no progress

This typically occur when Docker is not allocated enough memory to perform the migrations. Try modifying your Docker configuration to allow more RAM for each container, temporarily stop make_serve and run make initdb again.

make initdb complains about PostgreSQL Version

You built a Warehouse install some time ago and PostgreSQL has been updated. If you do not need the data in your databases, it might be best to just blow away your builds + docker containers and start again: make purge docker volume rm $(docker volume ls -q --filter dangling=true)

Compilation errors in non-Docker development

While Warehouse is designed to be developed in using Docker, you may have tried to install Warehouse’s requirements in your system or virtual environment. This is discouraged as it can result in compilation errors due to your system not including libraries or binaries required by some of Warehouse’s dependencies.

An example of such dependency is psycopg2 which requires PostgreSQL binaries and will fail if not present.

If there’s a specific use case you think requires development outside Docker please raise an issue in Warehouse’s issue tracker.

Disabling services locally

Some services, such as Elasticsearch, consume a lot of resources when running locally, but might not always be necessary when doing local development.

To disable these locally, you can create a docker-compose.override.yaml file to override any settings in the docker-compose.yaml file. To individually disable services, modify their entrypoint to do something else:

version: "3"

    entrypoint: ["echo", "Elasticsearch disabled"]

Note that disabling services might cause things to fail in unexpected ways.

This file is ignored in Warehouse’s .gitignore file, so it’s safe to keep in the root of your local repo.

Docker and Windows Subsystem for Linux Quirks

Once you have installed Docker for Windows, the Windows Subsystem for Linux, and Docker and Docker Compose in WSL, there are some extra configuration steps to deal with current quirks in WSL. Nick Janetakis has a detailed blog post on these steps, including installation, but this is a summary of the required steps:

1. In WSL, run sudo mkdir /c and sudo mount --bind /mnt/c /c to mount your root drive at /c (or whichever drive you are using). You should clone into this mount and run docker-compose from within it, to ensure that when volumes are linked into the container they can be found by Hyper-V.

2. In Windows, configure Docker to enable “Expose daemon on tcp://localhost:2375 without TLS”. Note that this may expose your machine to certain remote code execution attacks, so use with caution.

3. Add export DOCKER_HOST=tcp:// to your .bashrc file in WSL, and/or run it directly to enable for the current session. Without this, the docker command in WSL will not be able to find the daemon running in Windows.

Building Styles

Styles are written in the scss variant of Sass and compiled using gulp. They will be automatically built when changed when make serve is running.

Running the Interactive Shell

There is an interactive shell available in Warehouse which will automatically configure Warehouse and create a database session and make them available as variables in the interactive shell.

To run the interactive shell, simply run:

make shell

The interactive shell will have the following variables defined in it:


The Pyramid Configurator object which has already been configured by Warehouse.


The SQLAlchemy ORM Session object which has already been configured to connect to the database.

To use the db object in the interactive shell, import the class you’re planning to use. For example, if I wanted to use the User object, I would do this:

$ make shell
docker-compose run --rm web python -m warehouse shell
Starting warehouse_redis_1 ...
>>> from warehouse.accounts.models import User
>>> db.query(User).filter_by(username='test').all()

You can also run the IPython shell as the interactive shell. To do so export the environment variable WAREHOUSE_IPYTHON_SHELL prior to running the make build step:


Now you will be able to run the make shell command to get the IPython shell.

Running tests and linters


PostgreSQL 9.4 is required because of pgcrypto extension

The Warehouse tests are found in the tests/ directory and are designed to be run using make.

To run all tests, in the root of the repository:

make tests

This will run the tests with the supported interpreter as well as all of the additional testing that we require.

If you want to run a specific test, you can use the T variable:

T=tests/unit/i18n/ make tests

You can run linters, programs that check the code, with:

make lint

Warehouse uses black for opinionated formatting and linting. You can reformat with:

make reformat

Building documentation

The Warehouse documentation is stored in the docs/ directory. It is written in reStructured Text and rendered using Sphinx.

Use make to build the documentation. For example:

make docs

The HTML documentation index can now be found at docs/_build/html/index.html.

Building the docs requires Python 3.8. If it is not installed, the make command will give the following error message:

make: python3.8: Command not found
Makefile:53: recipe for target '.state/env/pyvenv.cfg' failed
make: *** [.state/env/pyvenv.cfg] Error 127

Building translations

Warehouse is translated into a number of different locales, which are stored in the warehouse/locale/ directory.

These translation files contain references to untranslated text in source code and HTML templates, as well as the translations which have been submitted by our volunteer translators.

When making changes to files with strings marked for translation, it’s necessary to update these references any time source strings are change, or the line numbers of the source strings in the source files.

Use make to build the translations. For example:

make translations

What next?

Look at our open issues that are labelled “good first issue”, find one you want to work on, comment on it to say you’re working on it, then submit a pull request. Use our Submitting patches documentation to help.

Talk with us

You can find us via a GitHub issue, #pypa or #pypa-dev on Libera, the PyPA Discord or Discourse, to ask questions or get involved. And you can meet us in person at packaging sprints.

Learn about Warehouse and packaging

Resources to help you learn Warehouse’s context: