Named Outputs

Tags: Basic

By default, Flyte names the outputs of a task or workflow using a standardized convention. All the outputs are named as o1, o2, o3, ... o<n>. where o is the standard prefix and 1, 2, .. <n> is the index position within the return values.

It is possible to give custom names to the outputs of a task or a workflow, so it’s easier to refer to them while debugging or visualising them in the UI. This is not possible natively in Python, so Flytekit provides an alternative using typing.NamedTuple.

The following example shows how to name outputs of a task and a workflow.

import typing

from flytekit import task, workflow

Named outputs can be declared inline as in the following task signature.

Note

Note that the name of the NamedTuple does not matter, but the names and types of the variables do. We used a a default name like OP. NamedTuples can be inline, but by convention we prefer to declare them, as pypy linter errors can be avoided this way.

def say_hello() -> typing.NamedTuple("OP", greet=str):
    pass

hello_output = typing.NamedTuple("OP", greet=str)


@task
def say_hello() -> hello_output:
    return hello_output("hello world")

You can also declare the NamedTuple ahead of time and then use it in the signature as follows:

wf_outputs = typing.NamedTuple("OP2", greet1=str, greet2=str)

As shown in this example, you can now refer to the declared NamedTuple. As seen in the workflow, say_hello returns a tuple. Like other tuples, you can simply unbox it inline. Also the workflow itself returns a tuple. You can also construct the tuple as you return the output.

Note

Note that we are de-referencing the individual task execution outputs because named-outputs use NamedTuple which are tuples that need to be de-referenced.

@workflow
def my_wf() -> wf_outputs:
    return wf_outputs(say_hello().greet, say_hello().greet)

The workflow can be executed as usual.

if __name__ == "__main__":
    print(f"Running my_wf() {my_wf()}")