Repositories

Repositories #

Poetry supports the use of PyPI and private repositories for discovery of packages as well as for publishing your projects.

By default, Poetry is configured to use the PyPI repository, for package installation and publishing.

So, when you add dependencies to your project, Poetry will assume they are available on PyPI.

This represents most cases and will likely be enough for most users.

Private Repository Example #

Installing from private package sources #

By default, Poetry discovers and installs packages from PyPI. But, you want to install a dependency to your project for a simple API repository? Let’s do it.

First, configure the package source as a supplemental (or explicit) package source to your project.

poetry source add --priority=supplemental foo https://pypi.example.org/simple/

Then, assuming the repository requires authentication, configure credentials for it.

poetry config http-basic.foo <username> <password>
Warning
Depending on your system configuration, credentials might be saved in your command line history. Many shells do not save commands to history when they are prefixed by a space character. For more information, please refer to your shell’s documentation.

Once this is done, you can add dependencies to your project from this source.

poetry add --source foo private-package

Publishing to a private repository #

Great, now all that is left is to publish your package. Assuming you’d want to share it privately with your team, you can configure the Upload API endpoint for your publishable repository.

poetry config repositories.foo https://pypi.example.org/legacy/
Note

If you need to use a different credential for your package source, then it is recommended to use a different name for your publishing repository.

poetry config repositories.foo-pub https://pypi.example.org/legacy/
poetry config http-basic.foo-pub <username> <password>

Now, all the is left is to build and publish your project using the publish.

poetry publish --build --repository foo-pub

Package Sources #

By default, Poetry is configured to use the Python ecosystem’s canonical package index PyPI.

Note
With the exception of the implicitly configured source for PyPI named pypi, package sources are local to a project and must be configured within the project’s pyproject.toml file. This is not the same configuration used when publishing a package.

Project Configuration #

These package sources may be managed using the source command for your project.

poetry source add foo https://foo.bar/simple/
Note
If your package source requires credentials or certificates, please refer to the relevant sections below.

This will generate the following configuration snippet in your pyproject.toml file.

[[tool.poetry.source]]
name = "foo"
url = "https://foo.bar/simple/"
priority = "primary"

If priority is undefined, the source is considered a primary source that takes precedence over PyPI, secondary, supplemental and explicit sources.

Package sources are considered in the following order:

  1. default source,
  2. primary sources,
  3. implicit PyPI (unless disabled by another default source or configured explicitly),
  4. secondary sources (DEPRECATED),
  5. supplemental sources.

Explicit sources are considered only for packages that explicitly indicate their source.

Within each priority class, package sources are considered in order of appearance in pyproject.toml.

Note

If you want to change the priority of PyPI, you can set it explicitly, e.g.

poetry source add --priority=primary PyPI

If you prefer to disable PyPI completely, you may choose to set one of your package sources to be the default or configure PyPI as explicit source.

Default Package Source #

By default, Poetry configures PyPI as the default package source for your project. You can alter this behaviour and exclusively look up packages only from the configured package sources by adding a single source with priority = "default".

poetry source add --priority=default foo https://foo.bar/simple/
Warning

In a future version of Poetry, PyPI will be disabled automatically if at least one custom primary source is configured. If you are using custom sources in addition to PyPI, you should configure PyPI explicitly with a certain priority, e.g.

poetry source add --priority=primary PyPI

This way, the priority of PyPI can be set in a fine-granular way.

The equivalent specification in pyproject.toml is:

[[tool.poetry.source]]
name = "pypi"
priority = "primary"

Omit the url when specifying PyPI explicitly. Because PyPI is internally configured with Poetry, the PyPI repository cannot be configured with a given URL. Remember, you can always use poetry check to ensure the validity of the pyproject.toml file.

Warning
Configuring a custom package source as default, will effectively disable PyPI as a package source for your project.

Secondary Package Sources (DEPRECATED) #

Deprecated in 1.5.0

If package sources are configured as secondary, all it means is that these will be given a lower priority when selecting compatible package distribution that also exists in your default and primary package sources. If the package source should instead be searched only if higher-priority repositories did not return results, please consider a supplemental source instead.

You can configure a package source as a secondary source with priority = "secondary" in your package source configuration.

poetry source add --priority=secondary https://foo.bar/simple/

There can be more than one secondary package source.

Warning
Secondary package sources are deprecated in favor of supplemental package sources.

Supplemental Package Sources #

Introduced in 1.5.0

Package sources configured as supplemental are only searched if no other (higher-priority) source yields a compatible package distribution. This is particularly convenient if the response time of the source is high and relatively few package distributions are to be fetched from this source.

You can configure a package source as a supplemental source with priority = "supplemental" in your package source configuration.

poetry source add --priority=supplemental https://foo.bar/simple/

There can be more than one supplemental package source.

Explicit Package Sources #

Introduced in 1.5.0

If package sources are configured as explicit, these sources are only searched when a package configuration explicitly indicates that it should be found on this package source.

You can configure a package source as an explicit source with priority = "explicit" in your package source configuration.

poetry source add --priority=explicit foo https://foo.bar/simple/

There can be more than one explicit package source.

Package Source Constraint #

All package sources (including secondary and possibly supplemental sources) will be searched during the package lookup process. These network requests will occur for all sources, regardless of if the package is found at one or more sources.

In order to limit the search for a specific package to a particular package repository, you can specify the source explicitly.

poetry add --source internal-pypi httpx

This results in the following configuration in pyproject.toml:

[tool.poetry.dependencies]
...
httpx = { version = "^0.22", source = "internal-pypi" }

[[tool.poetry.source]]
name = "internal-pypi"
url = ...
priority = ...
Note
A repository that is configured to be the only source for retrieving a certain package can itself have any priority. In particular, it does not need to have priority "explicit". If a repository is configured to be the source of a package, it will be the only source that is considered for that package and the repository priority will have no effect on the resolution.
Note

Package source keys are not inherited by their dependencies. In particular, if package-A is configured to be found in source = internal-pypi, and package-A depends on package-B that is also to be found on internal-pypi, then package-B needs to be configured as such in pyproject.toml. The easiest way to achieve this is to add package-B with a wildcard constraint:

poetry add --source internal-pypi package-B@*

This will ensure that package-B is searched only in the internal-pypi package source. The version constraints on package-B are derived from package-A (and other client packages), as usual.

If you want to avoid additional main dependencies, you can add package-B to a dedicated dependency group:

poetry add --group explicit --source internal-pypi package-B@*
Note
Package source constraints are strongly suggested for all packages that are expected to be provided only by one specific source to avoid dependency confusion attacks.

Supported Package Sources #

Python Package Index (PyPI) #

Poetry interacts with PyPI via its JSON API. This is used to retrieve a requested package’s versions, metadata, files, etc.

Note
If the the package’s published metadata is invalid, Poetry will download the available bdist/sdist to inspect it locally to identify the relevant metadata.

If you want to explicitly select a package from PyPI you can use the --source option with the add command, like shown below.

poetry add --source pypi httpx@^0.22.0

This will generate the following configuration snippet in your pyproject.toml file.

httpx = {version = "^0.22.0", source = "pypi"}
Warning
If any source within a project is configured with priority = "default", The implicit pypi source will be disabled and not used for any packages.

Simple API Repository #

Poetry can fetch and install package dependencies from public or private custom repositories that implement the simple repository API as described in PEP 503.

Warning
When using sources that distributes large wheels without providing file checksum in file URLs, Poetry will download each candidate wheel at least once in order to generate the checksum. This can manifest as long dependency resolution times when adding packages from this source.

These package sources may be configured via the following command in your project.

poetry source add testpypi https://test.pypi.org/simple/
Note
Note the trailing /simple/. This is important when configuring PEP 503 compliant package sources.

In addition to PEP 503, Poetry can also handle simple API repositories that implement PEP 658 (Introduced in 1.2.0). This is helpful in reducing dependency resolution time for packages from these sources as Poetry can avoid having to download each candidate distribution, in order to determine associated metadata.

Note

Why does Poetry insist on downloading all candidate distributions for all platforms when metadata is not available?

The need for this stems from the fact that Poetry’s lock file is platform-agnostic. This means, in order to resolve dependencies for a project, Poetry needs metadata for all platform specific distributions. And when this metadata is not readily available, downloading the distribution and inspecting it locally is the only remaining option.

Introduced in 1.2.0

Some projects choose to release their binary distributions via a single page link source that partially follows the structure of a package page in PEP 503.

These package sources may be configured via the following command in your project.

poetry source add jax https://storage.googleapis.com/jax-releases/jax_releases.html
Note
All caveats regarding slower resolution times described for simple API repositories do apply here as well.

Publishable Repositories #

Poetry treats repositories to which you publish packages as user specific and not project specific configuration unlike package sources. Poetry, today, only supports the Legacy Upload API when publishing your project.

These are configured using the config command, under the repositories key.

poetry config repositories.testpypi https://test.pypi.org/legacy/
Note
Legacy Upload API URLs are typically different to the same one provided by the repository for the simple API. You’ll note that in the example of Test PyPI, both the host (test.pypi.org) as well as the path (/legacy) are different to its simple API (https://test.pypi.org/simple).

Configuring Credentials #

If you want to store your credentials for a specific repository, you can do so easily:

poetry config http-basic.foo <username> <password>

If you do not specify the password you will be prompted to write it.

Note

To publish to PyPI, you can set your credentials for the repository named pypi.

Note that it is recommended to use API tokens when uploading packages to PyPI. Once you have created a new token, you can tell Poetry to use it:

poetry config pypi-token.pypi <my-token>

If you still want to use your username and password, you can do so with the following call to config.

poetry config http-basic.pypi <username> <password>

You can also specify the username and password when using the publish command with the --username and --password options.

If a system keyring is available and supported, the password is stored to and retrieved from the keyring. In the above example, the credential will be stored using the name poetry-repository-pypi. If access to keyring fails or is unsupported, this will fall back to writing the password to the auth.toml file along with the username.

Keyring support is enabled using the keyring library. For more information on supported backends refer to the library documentation.

Note

Poetry will fallback to Pip style use of keyring so that backends like Microsoft’s artifacts-keyring get a chance to retrieve valid credentials. It will need to be properly installed into Poetry’s virtualenv, preferably by installing a plugin.

If you are letting Poetry manage your virtual environments you will want a virtualenv seeder installed in Poetry’s virtualenv that installs the desired keyring backend during poetry install. To again use Azure DevOps as an example: azure-devops-artifacts-helpers provides such a seeder. This would of course best achieved by installing a Poetry plugin if it exists for you use case instead of doing it yourself.

Alternatively, you can use environment variables to provide the credentials:

export POETRY_PYPI_TOKEN_PYPI=my-token
export POETRY_HTTP_BASIC_PYPI_USERNAME=<username>
export POETRY_HTTP_BASIC_PYPI_PASSWORD=<password>

See Using environment variables for more information on how to configure Poetry with environment variables.

If your password starts with a dash (e.g. randomly generated tokens in a CI environment), it will be parsed as a command line option instead of a password. You can prevent this by adding double dashes to prevent any following argument from being parsed as an option.

poetry config -- http-basic.pypi myUsername -myPasswordStartingWithDash

Certificates #

Custom certificate authority and mutual TLS authentication #

Poetry supports repositories that are secured by a custom certificate authority as well as those that require certificate-based client authentication. The following will configure the “foo” repository to validate the repository’s certificate using a custom certificate authority and use a client certificate (note that these config variables do not both need to be set):

poetry config certificates.foo.cert /path/to/ca.pem
poetry config certificates.foo.client-cert /path/to/client.pem
Note

The value of certificates.<repository>.cert can be set to false if certificate verification is required to be skipped. This is useful for cases where a package source with self-signed certificates are used.

poetry config certificates.foo.cert false
Warning
Disabling certificate verification is not recommended as it is does not conform to security best practices.

Caches #

Poetry employs multiple caches for package sources in order to improve user experience and avoid duplicate network requests.

The first level cache is a Cache-Control header based cache for almost all HTTP requests.

Further, every HTTP backed package source caches metadata associated with a package once it is fetched or generated. Additionally, downloaded files (package distributions) are also cached.

Debugging Issues #

If you encounter issues with package sources, one of the simplest steps you might take to debug an issue is rerunning your command with the --no-cache flag.

poetry --no-cache add pycowsay

If this solves your issue, you can consider clearing your cache using the cache command.

Alternatively, you could also consider enabling very verbose logging -vvv along with the --no-cache to see network requests being made in the logs.