Plugin Concepts

Like the Pulp Core itself, all Pulp Plugins are Django Applications, and could be created like any other Django app with pulpcore-manager startapp <your_plugin>. However, instead of writing all of the boilerplate yourself, it is recommended that you start your plugin by utilizing the Plugin Template. This guide will assume that you have used the plugin_template, but if you are interested in the details of what it provides you, please see Plugin Django Application for more information for how plugins are “discovered” and connected to the pulpcore Django app. Additional information is given as inline comments in the template.

Plugin API Usage

Plugin Applications interact with pulpcore with two high level interfaces, subclassing and adding tasks.


Pulp Core and each plugin utilize Django and the Django Rest Framework. Each plugin provides Models, Serializers, and Viewsets. For each object that a plugin writer needs to make, the pulpcore.plugin API provides base classes. These base classes handle most of the boilerplate code, resulting in CRUD for each object out of the box.

Validating Models

Pulp ensures validity of its database models by carefully crafted serializers. So all instances where resources are created or updated, those serializers must be used.

To create a MyModel from a data dictionary, the MyModelSerializer can be used like:

serializer = MyModelSerializer(data=data)
instance = serializer.create(serializer.validated_data)

In the stages pipeline, you want to instantiate the content units without saving them to database right away. The ContentSaver stage will then persist the objects in the database. This can be established by:

# In MyPluginFirstStage::run
# <...>
serializer = MyModelSerializer(data=data)
d_content = DeclarativeContent(
await self.put(d_content)
# <...>


Any action that can run for a long time should be an asynchronous task. Plugin writers do not need to understand the internals of the pulpcore tasking system, workers automatically execute tasks from RQ, including tasks deployed by plugins.

Making Temporary Files Available to Tasks

Sometimes, files must be brought forward from a ViewSet to an executing task. The files may or may not end up being artifacts in the end. To tackle this, one should use PulpTemporaryFile.

# Example 1 - Saving a temporary file:
temp_file = PulpTemporaryFile(file=my_file)

# Example 2 - Validating the digest and saving a temporary file:
temp_file = PulpTemporaryFile.init_and_validate(
    my_file, expected_digests={'md5': '912ec803b2ce49e4a541068d495ab570'}

# Example 3 - Creating an Artifact from the PulpTemporaryFile:
    artifact = Artifact.from_pulp_temporary_file(temp_file)
except Exception:

When dealing with a clustered deployment, different pulp services are not guaranteed to share a common filesystem (like /usr/share/pulp). PulpTemporaryFile is the alternative for creating files with the same storage technology that the artifacts use. Therefore, the temporary files are accessible by all pulp instances.


The tasking system adds a concept called reservations which ensures that actions that act on the same resources are not run at the same time. To ensure data correctness, any action that alters the content of a repository (thus creating a new version) must be run asynchronously, locking on the repository and any other models which cannot change during the action. For example, sync tasks must be asynchronous and lock on the repository and the remote. Publish should lock on the repository version being published as well as the publisher.

Deploying Tasks

Tasks are deployed from Views or Viewsets, please see Kick off Tasks.

Task Groups

Sometimes, you may want to create many tasks to perform different parts of one larger piece of work, but you need a simple means to track the progress of these many tasks. Task Groups serve this purpose by providing details on the number of associated tasks in each possible state. For more details, please see Kick off Tasks.


GroupProgressReport can track progress of each task in that group. GroupProgressReport needs to be created and associated to the TaskGroup. From within a task that belongs to the TaskGroup, the GroupProgressReport needs to be updated.

# Once a TaskGroup is created, plugin writers should create GroupProgressReport objects
# ahead, so tasks can find them and update the progress.
task_group = TaskGroup(description="Migration Sub-tasks")
group_pr = GroupProgressReport(
    message="Repo migration",
# When a task that will be executing certain work, which is part of a TaskGroup, it will look
# for the TaskGroup it belongs to and find appropriate progress report by its code and will
# update it accordingly.
task_group = TaskGroup.current()
progress_repo = task_group.group_progress_reports.filter(code='create.repo_version')
progress_repo.update(done=F('done') + 1)
# To avoid race conditions/cache invalidation issues, this pattern needs to be used so that
# operations are performed directly inside the database:

# .update(done=F('done') + 1)

# See:
# Important: F() objects assigned to model fields persist after saving the model instance and
# will be applied on each save(). Do not use save() and use update() instead, otherwise
# refresh_from_db() should be called after each save()

Content Protection

Users can configure a ContentGuard to protect a Distribution on their own, but some plugins want to offer built-in content protection features. For example pulp_container may only want a user to download container images they have rights to based on some permissions system pulp_container could provide.

For more information, see the ContentGuard Usage by Plugin Writers documentation.

Plugin Settings

Plugins can define settings by creating a <your plugin>.app.settings module containing settings as you would define in the Django Settings File itself. pulpcore ships the actual file so settings cannot be added directly as with most Django deployments. Instead as each plugin is loaded, pulpcore looks for the <your plugin>.app.settings module and uses dynaconf to overlay the settings on top of pulpcore’s settings and user provided settings.

Settings are parsed in the following order with later settings overwriting earlier ones:

  1. Settings from /etc/pulp/

  2. Settings from (the pulpcore provided settings defaults).

  3. Plugin settings from <your plugin>.app.settings.

In some cases, a setting should not overwrite an existing setting, but instead add to it. For example, consider adding a custom log handler or logger to the LOGGING settings. You don’t want to fully overwrite it, but instead add or overwrite only a sub-portion. dynaconf provides the dynaconf_merge feature which is for merging settings instead of overwriting them. For example, pulp_ansible makes use of this here.

Custom API URL Routes

The typical plugin viewsets are all suburls under /pulp/api/v3/, but some content types require additional urls outside of this area. For example pulp_ansible provides the Galaxy API at /pulp_ansible/galaxy/.

Place a that defines a urlpatterns at the root of your Python package, and the pulpcore plugin loading code will append those urls to the url root. This allows your to be a typical Django file. For example pulp_ansible uses a defined here

Custom Content App Routes

The Content App may also require custom routes, for example pulp_container defines some. Read more about how to customize the content app with custom routes.

Configuring Reverse Proxy with Custom URLs

When a plugin requires either Pulp API or Pulp Content App custom urls, the reverse proxy, i.e. either Nginx or Apache, need to receive extra configuration snippets to know which service to route the custom URLs to.

A best practice is to document clearly the custom URL requirements your plugin needs. Although the installer can automatically install plugin snippets, other environments, e.g. k8s or docker or docker containers may still need to configure them manually. Having clear docs is a minimum.

You can ship webserver snippets as part of your Python package with three steps:

1. Create a python package named webserver_snippets directory inside your app, e.g. Like all Python packages it will have an

2. Create an nginx.conf and an apache.conf, and the installer will symlink to the correct one depending on which reverse proxy is installed. Please create both as the installer supports both.

3. Create an entry in to have the packaged plugin include the apache.conf and nginx.conf files.

Here is an example in pulp_ansible’s webserver configs.

For the nginx.conf you can use variables with the names pulp-api and pulp-content as the location for the backend services. For example, to route the url /pulp_ansible/galaxy/ to the Pulp API you could have your nginx.conf contain:

location /pulp_ansible/galaxy/ {
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto $scheme;
    proxy_set_header Host $http_host;
    # we don't want nginx trying to do something clever with
    # redirects, we set the Host: header above already.
    proxy_redirect off;
    proxy_pass http://pulp-api;

The Apache config provides variables containing the location of the Pulp Content App and the Pulp API as pulp-api and pulp-content respectively. Below is an equivalent snippet to the one above, only for Apache:

ProxyPass /pulp_ansible/galaxy http://${pulp-api}/pulp_ansible/galaxy
ProxyPassReverse /pulp_ansible/galaxy http://${pulp-api}/pulp_ansible/galaxy

For the entry, you’ll likely want one like the example below which was taken from pulp_ansible’s

include pulp_ansible/app/webserver_snippets/*

Overriding the Reverse Proxy Route Configuration

Sometimes a plugin may want to control the reverse proxy behavior of a URL at the webserver. For example, perhaps an additional header may want to be set at the reverse proxy when those urls are forwarded to the plugin’s Django code. To accomplish this, the custom app route can be used when it specifies a more-specific route than the installer’s base webserver configuration provides.

For example assume the header FOO should be set at the url /pulp/api/v3/foo_route. Below are two examples of a snippet that could do this (one for Nginx and another for Apache).

Nginx example:

location /pulp/api/v3/foo_route {
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto $scheme;
    proxy_set_header Host $http_host;

    proxy_set_header FOO 'asdf';  # This is the custom part

    # we don't want nginx trying to do something clever with
    # redirects, we set the Host: header above already.
    proxy_redirect off;
    proxy_pass http://pulp-api;

Apache example:

<Location "/pulp/api/v3/foo_route">
    ProxyPass /pulp/api http://${pulp-api}/pulp/api
    ProxyPassReverse /pulp/api http://${pulp-api}/pulp/api
    RequestHeader set FOO "asdf"

These snippets work because both Nginx and Apache match on “more-specific” routes first regardless of the order in the config file. The installer ships the a default of /pulp/api/v3 so anything containing another portion after v3 such as /pulp/api/v3/foo_route would be more specific.

Plugin API Stability and Deprecation Policy

The pulpcore.plugin API can introduce breaking changes, and will be introduced in the following way. For this example, assume that pulpcore 3.8 introduces a breaking change by changing the call signature of a method named def foo(a, b) which is importable via the plugin API.

In 3.8 the following changes happen:

  1. The new method would be introduced as a new named function def the_new_foo(...) or some similar name.

  2. The existing method signature def foo(a, b) is left in-tact.

  3. The foo method would have the a Python DeprecationWarning added to it such as:

    warnings.warn("foo() is deprecated and will be removed in pulpcore==3.9; use the_new_foo().", warnings.DeprecationWarning)
  4. A CHANGES/plugin_api/XXXX.deprecation changelog entry is created explaining how to port plugin code onto the new call interface.

Then in 3.9 the following happens:

  1. The def foo(a, b) method is deleted entirely.

  2. A CHANGES/plugin_api/XXXX.removal changelog entry is created explaining what has been removed.


Python DeprecationWarning log statements are shown to users of your plugin when using a deprecated call interface. This is by design to raise general awareness that the code in-use will eventually be removed.

This also applies to models importable from pulpcore.plugin.models. For example, an attribute that is being renamed or removed would follow a similar deprecation process described above to allow plugin code one release cycle to update their code compatibility.


It’s recommended to use the Pulp 3 Installer to install your plugin. Generally you can do this by configuring pulp_install_plugins variable with your Python package’s name. For example for pulp-file:

  pulp-file: {}

Custom Installation Tasks

Custom installation steps for a plugin can be added to the installer which are run only when your plugin is in the pulp_install_plugins configuration.

For example, pulp_rpm requires several system-level dependencies that cannot be received from PyPI. The installer delivers these dependencies at install time through the pulp_rpm_prerequisites role. That role ships with the installer itself.

It’s also possible to add custom install behaviors for developers too. For exampe, the galaxy_ng plugin desires their web UI to be built from source for devel installs. That occurs in a custom galaxy_ui.yml task in the installers pulp_devel role.

For help contributing or changing a plugin-specific installation, please reach out to the installer maintainers either through the developer mailing list ( or on Freenode in the developer channel, #pulp-dev.