StackTask

A basic workflow framework built using Django and Django-Rest-Framework to help automate basic Admin tasks within an OpenStack cluster.

Primarily built as user registration service that fits into the OpenStack ecosystem alongside Keystone, its purpose to fill functionality missing from Keystone. Ultimately it is just a framework with actions that are tied to an endpoint and can require certain data fields and perform actions via the OpenStack clients.

Useful for automating generic admin tasks that users might request but otherwise can't do without the admin role. Also allows automating the signup and creation of new users, but also allows such requests to require approval first if wanted. Due to issuing of uri+tokens for final steps of some actions, allows for a password submit/reset system as well.

Functionality:

The main workflow consists of three possible steps which can be executed at different points in time, depending on how the TaskView is defined.

The base use case is three stages:

• Recieve Request
  • Validate request data against action serializers.
  • If valid, setup Task to represent the request, and the Actions specified for that TaskView.
  • The service runs the pre_approve function on all actions which should do any self validation to mark the actions themselves as valid or invalid, and populating the nodes in the Task based on that.
• Admin Approval
  • An admin looks at the Task and its notes.
  • If they decide it is safe to approve, they do so.
    • If there are any invalid actions approval will do nothing until the action data is updated and initial validation is rerun.
  • The service runs the post_approve function on all actions.
  • If any of the actions require a Token to be issued and emailed for additional data such as a user password, then that will occur.
    • If no Token is required, the Task will run submit actions, and be marked as complete.
• Token Submit
  • User submits the Token data.
  • The service runs the submit function on all actions, passing along the Token data, normally a password.
    • The action will then complete with the given final data.
  • Task is marked as complete.

There are cases and TaskViews that auto-approve, and thus automatically do the middle step right after the first. There are also others which do not need a Token and thus run the submit step as part of the second, or even all three at once. The exact number of 'steps' and the time between them depends on the definition of the TaskView.

Actions themselves can also effectively do anything within the scope of those three stages, and there is even the ability to chain multiple actions together, and pass data along to other actions.

The points that are modular, or will be made more modular in future, are the TaskViews and the actions tied to them. Adding new actions is easy, and attaching them to existing TaskViews is as well. Adding new TaskViews is also fairly easy, but will be made more modular in future (see Future Plans).

Creation and management of Tasks, Tokens, and Notifications is not modular and is the framework around the defined Actions and TaskViews that handles how they are executed. This helps keep the way Actions are executed consistent and simpler to maintain, but does also allow Actions to run almost any logic within those consistent steps.

Admin Endpoints:

Endpoints for the management of tasks, tokens, and notifications. Most of these are limited by roles, and are for admin use only.

• ../v1/task - GET
  • A json containing all tasks.
    • This will be updated to take parameters to refine the list.
• ../v1/task/ - GET
  • Get details for a specific task.
• ../v1/task/ - PUT
  • Update a task and retrigger pre_approve.
• ../v1/task/ - POST
  • approve a task
• ../v1/token - GET
  • A json containing all tokens.
    • This will be updated to take parameters to refine the list.
• ../v1/token - POST
  • Reissue tokens for a given task.
• ../v1/token - DELETE
  • Delete all expired tokens.
• ../v1/token/ - GET
  • return a json describing the actions and required fields for the token.
• ../v1/token/ - POST
  • submit the token.
• ../v1/notification - GET
  • Get a list of all unacknowledged notifications.
• ../v1/notification - POST
  • Acknowledge a list of notifications.
• ../v1/notification/ - GET
  • Details on a specific notification.
• ../v1/notification/ - POST
  • Acknowledge a specific notification.

Filtering Tasks, Tokens, and Notifications

The task, token, and notification list endpoints can be filtered using a slight variant of the Django ORM filters.

This is done but sending a json with filters via HTTP parameters:

{'filters': {'fieldname': { 'operation': 'value'}}

Example:

{'filters': {'task_id': { 'exact': '842433bb-fa08-4fc1-8c3b-aa9904ceb370'}}

Possible field lookup operations: https://docs.djangoproject.com/en/1.8/ref/models/querysets/#id4

Default TaskView Endpoints:

Basic default endpoints for the TaskViews.

• ../v1/actions/CreateProject - GET
  • return a json describing the actions and required fields for the endpoint.
• ../v1/actions/CreateProject - POST
  • unauthenticated endpoint
  • for signup of new users/projects.
  • task requires manual approval, sends a uri+token for password setup after the project is created and setup.
    • create project
    • add admin user to project
    • setup basic networking if needed
    • create user on password submit
• ../v1/actions/InviteUser - GET
  • return a json describing the actions and required fields for the endpoint.
• ../v1/actions/InviteUser - POST
  • authenticated endpoint limited by role
  • auto-approved
  • add/invite a user to your project
  • adds an existing user with the selected role, or if non-existent user sends a uri+token to them for setup user before adding the role.
  • allows adding of users to own project without needing an admin role
• ../v1/actions/EditUser - GET
  • return a json describing the actions and required fields for the endpoint.
  • also returns a list of users that can be edited on your project.
• ../v1/actions/EditUser - POST
  • authenticated endpoint limited by role
  • auto-approved
  • add/remove roles from a user on your project
• ../v1/actions/ResetPassword - GET
  • return a json describing the actions and required fields for the endpoint.
• ../v1/actions/ResetPassword - POST
  • unauthenticated endpoint
  • auto-approved
  • issue a uri+token to user email to reset password

OpenStack Style TaskView Endpoints:

For ease of integration with OpenStack, these endpoints are setup to work and partly mimic the way similar ones would work in Keystone. They work and use the TaskViews, with some specical changes for certain required endpoints.

• ../v1/openstack/users - GET
  • Returns a list of users on your project, and their roles.
  • Also returns a list of pending user invites.
• ../v1/openstack/users - POST
  • See above, same as ../v1/actions/InviteUser - POST
• ../v1/openstack/users/<user_id> - GET
  • Get details on the given user, including their roles on your project.
• ../v1/openstack/users/<user_id> - DELETE
  • Used to cancel a pending user invite.
• ../v1/openstack/users/<user_id>/roles - GET
  • Returns a list of roles for the user on your project.
• ../v1/openstack/users/<user_id>/roles - PUT
  • Add roles to a user.
• ../v1/openstack/users/<user_id>/roles - DELETE
  • Remove roles from a user.
• ../v1/openstack/roles - GET
  • Returns a list of roles you are allowed to edit on your project.
• ../v1/openstack/forgotpassword - POST
  • Submit a username/email to have a password reset token emailed to it.

More API Documentation:

While in debug mode the service will supply online browsable documentation via Django REST Swagger.

This is viewable at:

• ../docs

Implementation Details:

Project Requirements:

The requirements for the service started as a system capable of taking requests for user sign up, waiting for approval, then on approval doing various setup and creation actions, followed by sending a uri+token to the user to set their password.

Creating a user directly in Keystone before approval also was to be avoided, and storing the password before approval was to be avoided.

Due to the steps involved, and the time between them, data had to be stored somewhere, and some representation of the request had to be stored as well. The ability to tie other actions and pieces of automation to the process also seemed useful and time saving considering the steps setting up a user might involve.

If that was the case, the system should ideally also have been modular enough to allow swapping of actions if circumstances changed, or new pieces of automation needed to be added or removed. Pushing as much logic to the concept of an 'action' seemed the ideal situation.

What is an Action?

Actions are a generic database model which knows what 'type' of action it is. On pulling the actions related to a Task from the database we wrap it into the appropriate class type which handles all the logic associated with that action type.

An Action is both a simple database representation of itself, and a more complex in memory class that handles all the logic around it.

Each action class has the functions "pre_approve", "post_approve", and "submit". These relate to stages of the approval process, and any python code can be executed in those functions, some of which should ideally be validation that the data passed makes sense.

Multiple actions can be chained together under one Task and will execute in the defined order. Actions can pass information along via an in memory cache/field on the task object, but that is only safe for the same stage of execution. Actions can also store data back to the database if their logic requires some info passed along to a later step of execution.

See actions.models for a good idea of Actions.

What is a Task?

A task is a top level model representation of the request. It wraps the request metadata, and based on the TaskView, will have actions associated with it.

See api.models.

What is a Token?

A token is a unique identifier linking to a task, so that anyone submitting the token will submit to the actions related to the task.

See api.models.

What is an TaskView

TaskViews are classes which extend the base TaskView class and use its imbuilt functions to process actions. They also have actions associated with them and the inbuilt functions from the base class are there to process and validate those against data coming in.

The TaskView will process incoming data and build it into a Task, and the related Action classes.

They are very simple to define as the inbuilt functions handle all the real logic, but defining which functions of those are called changes the view to create a task that either requires approval or auto-approves.

The base TaskView class has three functions:

• get
  • just a basic view function that by default returns list of actions, and their required fields for the action view.
• process_actions
  • needs to be called in the TaskView definition
  • A function to run the processing and validation of request data for actions.
  • Builds and returns the task object, or the validation errors.
• approve
  • Takes a task and approves it, running post_approve actions and issuing a token if needed.
  • Used only if no admin approval is needed for Tasks create by this TaskView.

See api.v1.tasks and look at the TaskView class to get a better idea.

For a more complex variant, look at api.v1.openstack to see some more unique TaskViews specific to certain endpoints we needed to mimic OpenStack functionality.

Development:

Packaging and Installing:

While this is a Django application, it does not follow the standard Django folder structure because of certain packaging requirements. As such the project does not have a manage.py file and must be installed via setup.py or pip (if an sdist is built) to access the manage.py funcationality.

Rather than a standard Django application, treat this as a more standard python application in this regard.

Once installed, all the normal manage.py functions can be called directly on the 'stacktask' commandline function.

Dev Environment:

Dev is mainly done within a virtualenv setup alongside a devstack deployment.

For more info, see:

DEVSTACK_GUIDE.md

Running tests:

We use tox to build a venv and run the tests. The same tests are also run for us in CI via jenkins.

Provided you have tox and its requirements installed running tests is very simple:

$ tox

Adding Actions:

Adding new actions is done by creating a new django app in the actions module and defining the action models and their serializers. Action must extend the BaseAction class as defined in the actions.models module. They also must add themselves to the global store of actions (see the bottom of existing models modules).

The documentation for this is mainly inline.

For examples of actions look in: action.models and actions.tenant_setup.models.

Adding TaskViews:

Ideally this will be made pluggable in future, but for now requires updating and adding a url entry in api.v1.urls and adding an additional TaskView linked to that url.

For examples see the classes CreateProject, InviteUser, and ResetPassword in the api.v1.tasks module, and api.v1.openstack.

Setup/Deployment:

Debian packaging

Debian packaging is done with dh-virtualenv (https://github.com/spotify/dh-virtualenv). When a package is built, a new virtualenv is created and populated using the contents of requirements.txt.

Package building setup: apt-get install build-essential devscripts dh-virtualenv

Incrementing package version: 'vim setup.py' and increment version parameter. 'dch -i' and write release notes, set version.

Build the package: dpkg-buildpackage -us -uc

Now a debian package has been built that will unpack a virtualenv containing stacktask and all dependencies in a self-contained package, so they do not conflict with other python packages on the system.

Puppet module

Then a puppet module will be able to install the debian package, setup a database, and run the service via nginx and uwsgi in the virtualenv.

Custom Email Templates

Custom email templates are placed in:

/etc/stacktask/templates/

This is so that adding personalised or deployment specific templates is kept outside of the scope of the service itself and managed by the deployer.

Future Plans:

Most future plans are around adding additional Actions to the service, but there will be some features that will require some refactoring.

While we are presently working with the Keystone V3 API, groups are not being used, but we intend to update the service to also manage and handle user groups. Managing Domains isn't really doable, but having the service be able to accept Domains, and multiple Domain back-ends is being planned.

Additional Actions we wish to add in the near future:

• Update Quota
  • Admin is required to do this
  • Allows users to request a quota increase and by requiring an admin to simply check, and confirm the request, will make the process faster.
  • Makes it effectively a quick 2 step process.
  • For eventual heirarchical-multi-tenancy this would ideally send these qouta increase requests to the parent for approval.
• Stand-alone Setup Network Action + TaskView
  • For users who missed or forgot the step at Project creation, and want a quick network setup.
• Blank Slate
  • Doesn't need admin, and could reuse your own token.
  • Clear my entire project of any and all resources.
  • Already doable via the APIs, but would be nice to have it in an easy to request action so clients don't need to code it themselves.

Additional Actions for our own deployment (may be useful to others):

• openERP client create
  • An Action that will create a client in openERP and link them to the Project
  • Or if the Client already exists, just link them to the project.
    • Will need to automatically validate that the client doesn't already exist.

Features that might require a slight refactor:

• Break out the TaskViews into a set of explorable endpoints so adding and removing TaskViews is easier.
• Add optional serializers for token data.

Even less likely, and further far-future additions:

• Split the system into the api, a queue, and workers. That way tasks are processed asynchronously by the workers.
  • Will require a bunch of rethinking, but most of the core logic will be reused, with the workers simply waiting for events and executing them on the tasks/actions in much the same way as they are presently.
• Remove concept of predefined action steps entirely, setup Actions to have any possible number of 'steps'.
  • Will require moving actions to an iterator style pattern with a "next_action" style function as the driving force.
  • Will alter how chaining actions together works, thus may require a lot of work to define a sensible pattern for chaining them together.