GitHub Actions
We created prefix-dev/setup-pixi to facilitate using pixi in CI.
Usage#
- uses: prefix-dev/setup-pixi@v0.6.0
with:
pixi-version: v0.21.1
cache: true
auth-host: prefix.dev
auth-token: ${{ secrets.PREFIX_DEV_TOKEN }}
- run: pixi run test
Pin your action versions
Since pixi is not yet stable, the API of this action may change between minor versions.
Please pin the versions of this action to a specific version (i.e., prefix-dev/setup-pixi@v0.6.0
) to avoid breaking changes.
You can automatically update the version of this action by using Dependabot.
Put the following in your .github/dependabot.yml
file to enable Dependabot for your GitHub Actions:
version: 2
updates:
- package-ecosystem: github-actions
directory: /
schedule:
interval: monthly # (1)!
groups:
dependencies:
patterns:
- "*"
- or
daily
,weekly
Features#
To see all available input arguments, see the action.yml
file in setup-pixi
.
The most important features are described below.
Caching#
The action supports caching of the pixi environment.
By default, caching is enabled if a pixi.lock
file is present.
It will then use the pixi.lock
file to generate a hash of the environment and cache it.
If the cache is hit, the action will skip the installation and use the cached environment.
You can specify the behavior by setting the cache
input argument.
Customize your cache key
If you need to customize your cache-key, you can use the cache-key
input argument.
This will be the prefix of the cache key. The full cache key will be <cache-key><conda-arch>-<hash>
.
Only save caches on main
In order to not exceed the 10 GB cache size limit as fast, you might want to restrict when the cache is saved.
This can be done by setting the cache-write
argument.
Multiple environments#
With pixi, you can create multiple environments for different requirements.
You can also specify which environment(s) you want to install by setting the environments
input argument.
This will install all environments that are specified and cache them.
[project]
name = "my-package"
channels = ["conda-forge"]
platforms = ["linux-64"]
[dependencies]
python = ">=3.11"
pip = "*"
polars = ">=0.14.24,<0.21"
[feature.py311.dependencies]
python = "3.11.*"
[feature.py312.dependencies]
python = "3.12.*"
[environments]
py311 = ["py311"]
py312 = ["py312"]
Multiple environments using a matrix#
The following example will install the py311
and py312
environments in different jobs.
test:
runs-on: ubuntu-latest
strategy:
matrix:
environment: [py311, py312]
steps:
- uses: actions/checkout@v4
- uses: prefix-dev/setup-pixi@v0.6.0
with:
environments: ${{ matrix.environment }}
Install multiple environments in one job#
The following example will install both the py311
and the py312
environment on the runner.
- uses: prefix-dev/setup-pixi@v0.6.0
with:
environments: >- # (1)!
py311
py312
- run: |
pixi run -e py311 test
pixi run -e py312 test
-
separated by spaces, equivalent to
Caching behavior if you don't specify environments
If you don't specify any environment, the default
environment will be installed and cached, even if you use other environments.
Authentication#
There are currently three ways to authenticate with pixi:
- using a token
- using a username and password
- using a conda-token
For more information, see Authentication.
Handle secrets with care
Please only store sensitive information using GitHub secrets. Do not store them in your repository.
When your sensitive information is stored in a GitHub secret, you can access it using the ${{ secrets.SECRET_NAME }}
syntax.
These secrets will always be masked in the logs.
Token#
Specify the token using the auth-token
input argument.
This form of authentication (bearer token in the request headers) is mainly used at prefix.dev.
- uses: prefix-dev/setup-pixi@v0.6.0
with:
auth-host: prefix.dev
auth-token: ${{ secrets.PREFIX_DEV_TOKEN }}
Username and password#
Specify the username and password using the auth-username
and auth-password
input arguments.
This form of authentication (HTTP Basic Auth) is used in some enterprise environments with artifactory for example.
- uses: prefix-dev/setup-pixi@v0.6.0
with:
auth-host: custom-artifactory.com
auth-username: ${{ secrets.PIXI_USERNAME }}
auth-password: ${{ secrets.PIXI_PASSWORD }}
Conda-token#
Specify the conda-token using the conda-token
input argument.
This form of authentication (token is encoded in URL: https://my-quetz-instance.com/t/<token>/get/custom-channel
) is used at anaconda.org or with quetz instances.
- uses: prefix-dev/setup-pixi@v0.6.0
with:
auth-host: anaconda.org # (1)!
conda-token: ${{ secrets.CONDA_TOKEN }}
- or my-quetz-instance.com
Custom shell wrapper#
setup-pixi
allows you to run command inside of the pixi environment by specifying a custom shell wrapper with shell: pixi run bash -e {0}
.
This can be useful if you want to run commands inside of the pixi environment, but don't want to use the pixi run
command for each command.
- everything here will be run inside of the pixi environment
You can even run Python scripts like this:
- everything here will be run inside of the pixi environment
If you want to use PowerShell, you need to specify -Command
as well.
- run: | # (1)!
python --version | Select-String "3.11"
shell: pixi run pwsh -Command {0} # pwsh works on all platforms
- everything here will be run inside of the pixi environment
How does it work under the hood?
Under the hood, the shell: xyz {0}
option is implemented by creating a temporary script file and calling xyz
with that script file as an argument.
This file does not have the executable bit set, so you cannot use shell: pixi run {0}
directly but instead have to use shell: pixi run bash {0}
.
There are some custom shells provided by GitHub that have slightly different behavior, see jobs.<job_id>.steps[*].shell
in the documentation.
See the official documentation and ADR 0277 for more information about how the shell:
input works in GitHub Actions.
--frozen
and --locked
#
You can specify whether setup-pixi
should run pixi install --frozen
or pixi install --locked
depending on the frozen
or the locked
input argument.
See the official documentation for more information about the --frozen
and --locked
flags.
If you don't specify anything, the default behavior is to run pixi install --locked
if a pixi.lock
file is present and pixi install
otherwise.
Debugging#
There are two types of debug logging that you can enable.
Debug logging of the action#
The first one is the debug logging of the action itself. This can be enabled by for the action by re-running the action in debug mode:
Debug logging documentation
For more information about debug logging in GitHub Actions, see the official documentation.
Debug logging of pixi#
The second type is the debug logging of the pixi executable.
This can be specified by setting the log-level
input.
- One of
q
,default
,v
,vv
, orvvv
.
If nothing is specified, log-level
will default to default
or vv
depending on if debug logging is enabled for the action.
Self-hosted runners#
On self-hosted runners, it may happen that some files are persisted between jobs.
This can lead to problems or secrets getting leaked between job runs.
To avoid this, you can use the post-cleanup
input to specify the post cleanup behavior of the action (i.e., what happens after all your commands have been executed).
If you set post-cleanup
to true
, the action will delete the following files:
.pixi
environment- the pixi binary
- the rattler cache
- other rattler files in
~/.rattler
If nothing is specified, post-cleanup
will default to true
.
On self-hosted runners, you also might want to alter the default pixi install location to a temporary location. You can use pixi-bin-path: ${{ runner.temp }}/bin/pixi
to do this.
- uses: prefix-dev/setup-pixi@v0.6.0
with:
post-cleanup: true
pixi-bin-path: ${{ runner.temp }}/bin/pixi # (1)!
${{ runner.temp }}\Scripts\pixi.exe
on Windows
You can also use a preinstalled local version of pixi on the runner by not setting any of the pixi-version
,
pixi-url
or pixi-bin-path
inputs. This action will then try to find a local version of pixi in the runner's PATH.
Using the pyproject.toml
as a manifest file for pixi.#
setup-pixi
will automatically pick up the pyproject.toml
if it contains a [tool.pixi.project]
section and no pixi.toml
.
This can be overwritten by setting the manifest-path
input argument.
More examples#
If you want to see more examples, you can take a look at the GitHub Workflows of the setup-pixi
repository.