Your first workflow

Welcome to the WDL quickstart guide 🚀! This brief guide gives an overview of all the important bits of the WDL language to ensure you become an expert in writing your own tasks and workflows. Before diving into an example, let's take a closer look at the major concepts within the language.

Tasks

Tasks are the atomic units of computation within WDL. Tasks are comprised of a set of inputs (defined within the input section), a set of outputs (defined within the output section), and a command (defined within the command section), which is simply a Bash script to execute. Tasks also support defining requirements (defined within the requirements section) which dictate certain aspects of the task runtime environment. Importantly, tasks are executed within a container to ensure portability across different execution environments (e.g., your local computer, an HPC, or the cloud).

Workflows

Workflows string together tasks via their inputs and outputs into a larger computation graph that can be executed. Workflows are arranged similarly to tasks except that (a) they don't have a command or requirements section and (b) they make available more control flow facilities that aren't relevant within a task context. For example, the workflow below uses both conditional execution (the if statement) and a scatter-gather (the scatter keyword and messages output). Notably, workflows can also define inputs and outputs, and these generally serve as the global inputs and outputs for the execution of a workflow.

Inputs and Outputs

JSON is used for specifying both inputs to and outputs from a workflow. In the code block below the workflow, you can see the inputs that are specified for the top-level name parameter. You can also optionally provide a value for the is_pirate parameter. Further, the output of the workflow is communicated back to you via JSON. Executing workflows typically involves preparing the needed JSON files and reading the outputs within the JSON returned from your execution engine.

An example

For example, assume you wanted to write a task that greets someone, as defined in the name input, in multiple different languages. If that individual is a pirate, you might even wish to greet them conditionally with Ahoy! WDL allows you to express this in a straightforward manner by (a) constructing the atomic computation you'd like to achieve using a task and (b) running that task for each of your greetings using a workflow.

wdlCopy
version 1.3

task say_hello {
    input {
        String greeting
        String name
    }

    command <<<
        echo "~{greeting}, ~{name}!"
    >>>

    output {
        String message = read_string(stdout())
    }

    requirements {
        container: "ubuntu:latest"
    }
}

workflow main {
    input {
        String name
        Boolean is_pirate = false
    }

    Array[String] greetings = select_all([
        "Hello",
        "Hallo",
        "Hej",
        (
            if is_pirate
            then "Ahoy"
            else None
        ),
    ])

    scatter (greeting in greetings) {
        call say_hello {
            greeting,
            name,
        }
    }

    output {
        Array[String] messages = say_hello.message
    }
}

Running the example

If you were to run the example above with these inputs,

jsonCopy
{
  "main.name": "world",
  // "main.is_pirate": true,
}

the output of the above workflow would be the following.

jsonCopy
{
  "messages": [
    "Hello, world!",
    "Hallo, world!",
    "Hej, world!",
    // "Ahoy, world!" is included if `is_pirate` is set to `true` above.
  ]
}

Conclusion

This workflow, though simple, demonstrates the how WDL accomplishes its main values: namely, its human-readable/writable style and its straightforward but powerful control flow abstractions. You can learn more about the values of the WDL language on the Overview page.