Quick Start¶

Create and run your first Stardag task in 5 minutes.

Hello World¶

The simplest Stardag task is a function decorated with @sd.task:

import stardag as sd

@sd.task
def hello() -> str:
    return "Hello, Stardag!"

# Create a task instance
task = hello()

# Build (execute) the task
sd.build(task)

# Load the result
print(task.output().load())  # "Hello, Stardag!"

Let's break down what happened:

1. Task Definition¶

import stardag as sd

@sd.task
def hello() -> str:
    return "Hello, Stardag!"

The @sd.task decorator transforms your function into a task class. The return type annotation (-> str) tells Stardag how to serialize the output.

2. Task Instantiation¶

task = hello()

This creates a task instance - a declarative specification of what to compute. No computation happens yet.

3. Building¶

sd.build(task)

build() executes the task and saves the output to a deterministic file path.

4. Loading Results¶

task.output().load()

Retrieve the persisted result from storage.

Task with Parameters¶

Tasks can have parameters:

@sd.task
def greet(name: str) -> str:
    return f"Hello, {name}!"

task = greet(name="World")
sd.build(task)
print(task.output().load())  # "Hello, World!"

Key Concepts¶

Declarative: Task instances describe what to compute, not when.
Persistent: Results are saved to remote or local filesystem automatically (by default).
Deterministic: Output URIs are determined by task parameters; task instances are pointers to assets which we can reason about (know the location of) before they are materialized.

Inspecting Tasks¶

Tasks are Pydantic models with useful properties:

task = greet(name="World")

# View the task specification
print(repr(task))
# greet(version=None, name='World')

# Check if already complete
print(task.complete())
# False (before build), True (after build)

# View output path
print(task.output().uri)
# /path/to/.stardag/target-roots/default/greet/ab/cd/abcd1234....json

What's Next?¶

Continue to Your First DAG to learn how to create tasks with dependencies.