author Brian Hackett <bhackett1024@gmail.com>
Wed, 14 Nov 2018 16:09:58 -1000
changeset 446931 1c7fc8389e012c987347efefca6b35f3948b742a
parent 436909 bdfd4802db0cfb656dbb001676fdbe9766112b30
child 449374 72633f72bd41bff1edd3cdc3d870b10c869b3b22
permissions -rw-r--r--
Bug 1507359 Part 2 - Bindings and internal changes to allow ReplayDebugger to control child pausing/resuming, r=mccr8.


This document shows how to define an action in-tree such that it shows up in
supported user interfaces like Treeherder. For details on interface between
in-tree logic and external user interfaces, see `the actions.json spec`_.

At a very high level, the process looks like this:

 * The decision task produces an artifact, ``public/actions.json``, indicating
   what actions are available.

 * A user interface (for example, Treeherder or the Taskcluster tools) consults
   ``actions.json`` and presents appropriate choices to the user, if necessary
   gathering additional data from the user, such as the number of times to
   re-trigger a test case.

 * The user interface follows the action description to carry out the action.
   In most cases (``action.kind == 'task'``), that entails creating an "action
   task", including the provided information. That action task is responsible
   for carrying out the named action, and may create new sub-tasks if necessary
   (for example, to re-trigger a task).

Defining Action Tasks

There are two options for defining actions: creating a callback action, or
creating a custom action task.  A callback action automatically defines an
action task that will invoke a Python function of your devising.

A custom action task is an arbitrary task definition that will be created
directly.  In cases where the callback would simply call ``queue.createTask``,
a custom action task can be more efficient.

Creating a Callback Action

.. note:

    You can generate ``actions.json`` on the command line with ``./mach taskgraph actions``.

A *callback action* is an action that calls back into in-tree logic. That is,
you register the action with name, title, description, context, input schema and a
python callback. When the action is triggered in a user interface,
input matching the schema is collected, passed to a new task which then calls
your python callback, enabling it to do pretty much anything it wants to.

To create a new callback action you must create a file
``taskcluster/taskgraph/actions/my-action.py``, that at minimum contains::

  from registry import register_callback_action

      title='Say Hello',
      symbol='hw',  # Show the callback task in treeherder as 'hw'
      description="Simple **proof-of-concept** callback action",
      order=10000,  # Order in which it should appear relative to other actions
  def hello_world_action(parameters, graph_config, input, task_group_id, task_id, task):
      print "Hello was triggered from taskGroupId: " + taskGroupId

The arguments are:
  an instance of ``taskgraph.parameters.Parameters``, carrying decision task parameters from the original decision task.
  an instance of ``taskgraph.config.GraphConfig``, carrying configuration for this tree
  the input from the user triggering the action (if any)
  the target task group on which this action should operate
  the target task on which this action should operate (or None if it is operating on the whole group)
  the definition of the target task (or None, as for ``task_id``)

The example above defines an action that is available in the context-menu for
the entire task-group (result-set or push in Treeherder terminology). To create
an action that shows up in the context menu for a task we would specify the
``context`` parameter.

The ``order`` value is the sort key defining the order of actions in the
resulting ``actions.json`` file.  If multiple actions have the same name and
match the same task, the action with the smallest ``order`` will be used.

Setting the Action Context
The context parameter should be a list of tag-sets, such as
``context=[{"platform": "linux"}]``, which will make the task show up in the
context-menu for any task with ``task.tags.platform = 'linux'``. Below is
some examples of context parameters and the resulting conditions on
``task.tags`` (tags used below are just illustrative).

``context=[{"platform": "linux"}]``:
  Requires ``task.tags.platform = 'linux'``.
``context=[{"kind": "test", "platform": "linux"}]``:
  Requires ``task.tags.platform = 'linux'`` **and** ``task.tags.kind = 'test'``.
``context=[{"kind": "test"}, {"platform": "linux"}]``:
  Requires ``task.tags.platform = 'linux'`` **or** ``task.tags.kind = 'test'``.
  Requires nothing and the action will show up in the context menu for all tasks.
  Is the same as not setting the context parameter, which will make the action
  show up in the context menu for the task-group.
  (i.e., the action is not specific to some task)

The example action below will be shown in the context-menu for tasks with
``task.tags.platform = 'linux'``::

  from registry import register_callback_action

      symbol='re-c',  # Show the callback task in treeherder as 're-c'
      description="Create a clone of the task",
      context=[{'platform': 'linux'}]
  def retrigger_action(parameters, graph_config, input, task_group_id, task_id, task):
      # input will be None
      print "Retriggering: {}".format(task_id)
      print "task definition: {}".format(task)

When the ``context`` parameter is set, the ``task_id`` and ``task`` parameters
will provided to the callback. In this case the ``task_id`` and ``task``
parameters will be the ``taskId`` and *task definition* of the task from whose
context-menu the action was triggered.

Typically, the ``context`` parameter is used for actions that operate on
tasks, such as retriggering, running a specific test case, creating a loaner,
bisection, etc. You can think of the context as a place the action should
appear, but it's also very much a form of input the action can use.

Specifying an Input Schema
In call examples so far the ``input`` parameter for the callbacks has been
``None``. To make an action that takes input you must specify an input schema.
This is done by passing a JSON schema as the ``schema`` parameter.

When designing a schema for the input it is important to exploit as many of the
JSON schema validation features as reasonably possible. Furthermore, it is
*strongly* encouraged that the ``title`` and ``description`` properties in
JSON schemas is used to provide a detailed explanation of what the input
value will do. Authors can reasonably expect JSON schema ``description``
properties to be rendered as markdown before being presented.

The example below illustrates how to specify an input schema. Notice that while
this example doesn't specify a ``context`` it is perfectly legal to specify
both ``input`` and ``context``::

  from registry import register_callback_action

      title='Run All Tasks',
      symbol='ra-c',  # Show the callback task in treeherder as 'ra-c'
      description="**Run all tasks** that have been _optimized_ away.",
          'title': 'Action Options',
          'description': 'Options for how you wish to run all tasks',
          'properties': {
              'priority': {
                  'title': 'priority'
                  'description': 'Priority that should be given to the tasks',
                  'type': 'string',
                  'enum': ['low', 'normal', 'high'],
                  'default': 'low',
              'runTalos': {
                  'title': 'Run Talos'
                  'description': 'Do you wish to also include talos tasks?',
                  'type': 'boolean',
                  'default': 'false',
          'required': ['priority', 'runTalos'],
          'additionalProperties': False,
  def retrigger_action(parameters, graph_config, input, task_group_id, task_id, task):
      print "Create all pruned tasks with priority: {}".format(input['priority'])
      if input['runTalos']:
          print "Also running talos jobs..."

When the ``schema`` parameter is given the callback will always be called with
an ``input`` parameter that satisfies the previously given JSON schema.
It is encouraged to set ``additionalProperties: false``, as well as specifying
all properties as ``required`` in the JSON schema. Furthermore, it's good
practice to provide ``default`` values for properties, as user interface generators
will often take advantage of such properties.

It is possible to specify the ``schema`` parameter as a callable that returns
the JSON schema. It will be called with a keyword parameter ``graph_config``
with the `graph configuration <taskgraph-graph-config>` of the current

Once you have specified input and context as applicable for your action you can
do pretty much anything you want from within your callback. Whether you want
to create one or more tasks or run a specific piece of code like a test.

Conditional Availability

The decision parameters ``taskgraph.parameters.Parameters`` passed to
the callback are also available when the decision task generates the list of
actions to be displayed in the user interface. When registering an action
callback the ``availability`` option can be used to specify a callable
which, given the decision parameters, determines if the action should be available.
The feature is illustrated below::

  from registry import register_callback_action

      title='Say Hello',
      symbol='hw',  # Show the callback task in treeherder as 'hw'
      description="Simple **proof-of-concept** callback action",
      # Define an action that is only included if this is a push to try
      available=lambda parameters: parameters.get('project', None) == 'try',
  def try_only_action(parameters, graph_config, input, task_group_id, task_id, task):
      print "My try-only action"

Properties of ``parameters``  are documented in the
:doc:`parameters section <parameters>`. You can also examine the
``parameters.yml`` artifact created by decisions tasks.

Context can be similarly conditionalized by passing a function which returns
the appropriate context::

    context=lambda params:
        [{}] if int(params['level']) < 3 else [{'worker-implementation': 'docker-worker'}],

Creating Tasks

The ``create_tasks`` utility function provides a full-featured way to create
new tasks.  Its features include creating prerequisite tasks, operating in a
"testing" mode with ``./mach taskgraph action-callback --test``, and generating
artifacts that can be used by later action tasks to figure out what happened.
See the source for more detailed docmentation.

The artifacts are:

``task-graph.json`` (or ``task-graph-<suffix>.json``:
  The graph of all tasks created by the action task. Includes tasks
  created to satisfy requirements.
``to-run.json`` (or ``to-run-<suffix>.json``:
  The set of tasks that the action task requested to build. This does not
  include the requirements.
``label-to-taskid.json`` (or ``label-to-taskid-<suffix>.json``:
  This is the mapping from label to ``taskid`` for all tasks involved in
  the task-graph. This includes dependencies.

More Information

For further details on actions in general, see `the actions.json spec`_.
The hooks used for in-tree actions are set up by `ci-admin`_ based on configuration in `ci-configuration`_.

.. _the actions.json spec: https://docs.taskcluster.net/manual/tasks/actions/spec
.. _ci-admin: http://hg.mozilla.org/build/ci-admin/
.. _ci-configuration: http://hg.mozilla.org/build/ci-configuration/