Creating a Simple Task Graph

This tutorial will walk you through the process of creating a simple task graph.

We’ll create the scaffolding for Taskgraph and set up a single “Hello World” task. Once this is working, you will have the tools and knowledge to start adding your own tasks, and modifying configuration to meet your repository’s needs.

For the purposes of this tutorial, we’ll assume your repository is called myrepo.

Create the File System Layout

Create a file system layout similar to:

├── ci
│   ├── config.yml
│   └── hello
|       └── kind.yml
└── myrepo_taskgraph
    └── transforms

The taskcluster directory should appear at the root of the repository. It is possible to rename / put it somewhere else, but you’ll need to jump through a couple extra steps in that case. The location of the transforms file is similarly only a convention (but it must be importable).

Populate the Config File

The config file lives at taskcluster/ci/config.yml and must conform to the graph_config_schema.

Create the config.yml file to look like:

trust-domain: myrepo
task-priority: low
    # Values here depend on your Taskcluster instance's configuration.
            provisioner: '{trust-domain}-provisioner'
            implementation: docker-worker
            os: linux
            worker-type: '{trust-domain}-linux'
            name: myrepo

Here’s an explanation of the required top-level keys:

  • trust-domain - Trust domains help prevent permissions and worker pools from one project from being re-used in another. The value you should use here will be configured by your Taskcluster administrator. But we’ll use myrepo for the purposes of this tutorial.

  • task-priority - The priority of tasks for this repo.

  • workers - The set of worker pools available to the project. This will also be configured by your Taskcluster administrator.

  • taskgraph - A dict containing some miscellaneous configuration needed by Taskgraph.

Define a Task

Kinds are groupings of tasks that share certain characteristics with one another. Each subdirectory in taskcluster/ci corresponds to a different kind and contains a kind.yml file. These files define some properties that control how Taskgraph will generate the tasks, as well as the starting definitions of the tasks themselves. If you followed the layout above, you have a hello kind. For this next section we’ll be editing taskcluster/ci/hello/kind.yml.

  1. First declare a loader. Loaders determine how the task definitions get read. The most common is the transform loader:

    loader: taskgraph.loader.transform:loader
  2. Next declare the set of transforms that will be applied to tasks. Usually there is at least a kind specific set of transforms, as well as the general purpose -taskgraph.transforms.task transforms. Practically every task should use the latter, as they perform the final steps to modify the tasks into the format Taskcluster expects. In our example:

        - myrepo_taskgraph.transforms.hello:transforms
        - taskgraph.transforms.task:transforms
  3. Finally we define the task under the tasks key. The format for the initial definition here can vary wildly from one kind to another, it all depends on the transforms that are used. It’s conventional for transforms to define a schema (but not required). So often you can look at the first transform file to see what schema is expected of your job. But since we haven’t created the first transforms yet, let’s define our task like this for now:

            description: "Says hello to Taskcluster"
            text: "Taskcluster!"

Here is the combined kind.yml file:

loader: taskgraph.loader.transform:loader
    - myrepo_taskgraph.transforms.hello:transforms
    - taskgraph.transforms.task:transforms
        description: "Says hello to Taskcluster"
        text: "Taskcluster!"

Create the Transform

Transforms are Python generators that take a -taskgraph.transforms.base.TransformConfig instance and a generator that yields task definitions (in dictionary form) as input. It yields task definitions (which may or may not be modified) from the original inputs.

Typically transform files contain a schema, followed by one or more transform functions. Rather than break it down step by step, here’s what our taskcluster/myrepo_taskgraph/transforms/ file will look like (see comments for explanations):

from voluptuous import Optional, Required

from taskgraph.transforms.base import TransformSequence
from taskgraph.util.schema import Schema

# Define the schema. We use the `voluptuous` package to handle validation.
hello_description_schema = Schema({
    Required("text"): str,
    Optional("description"): str,

# Create a 'TransformSequence' instance. This class collects transform
# functions to run later.
transforms = TransformSequence()

# First let's validate tasks against the schema.

# Register our first transform functions via decorator.
def set_command(config, tasks):
    """Builds the command the task will run."""
    for task in tasks:
        task["command"] = f"bash -cx 'echo Hello {task.pop('text')}'"
        yield task

def build_task_description(config, tasks):
    """Sets the attributes required by transforms in
    for task in tasks:
        if "description" not in task:
            task["description"] = f"Says Hello {task['text']}"
        task["label"] = f"{config.kind}-{task.pop('name')}"
        # This is what was defined in `taskcluster/ci/config.yml`.
        task["worker-type"] = "linux"
        task["worker"] = {
            "command": task.pop["command"],
            "docker-image": "ubuntu:latest",
            "max-run-time": 300,  # seconds
        yield task

Generate the Taskgraph

Now it’s time to see if everything works! If you haven’t done so already, follow the Installation docs to install Taskgraph.

Next run the following command at the root of your repo:

taskgraph full

If all goes well, you should see some log output followed by a single task called hello-taskcluster. Try adding a second task to your tasks key in the kind.yml file and re-generating the graph. You should see both task labels!

Now run:

taskgraph morphed -J

The -J/--json flag will display the full JSON definition of your task. Morphed is the final phase of graph generation, so represents your task’s final form before it would get submitted to Taskcluster. In fact, if we hadn’t made up the trust domain and worker pool in config.yml, you could even copy / paste this definition into Taskcluster’s task creator!

Next you can check out the Submitting your Graph to Taskcluster tutorial or learn more about generating the taskgraph locally.