Skip to main content
Flytekit provides conditional as a first-class construct for selectively executing branches in a workflow. Conditions can test static values, workflow inputs, or outputs from upstream tasks.
Conditionals are restricted to binary and logical operators on primitive values. They are type-safe and highly performant because Flyte evaluates them without spawning additional pods.

​
How conditionals work

The pattern is: 
result = (
    conditional("branch-name")
    .if_(some_condition)
    .then(task_a(...))
    .elif_(other_condition)
    .then(task_b(...))
    .else_()
    .fail("No condition matched")
)
Every conditional must be complete — all possible branches must be accounted for. If no branch matches, the workflow fails unless you provide .else_().fail(...) or a default .else_().then(...).
Conditionals use bitwise & and | instead of Python’s and / or. Due to Python’s PEP-335, and, or, and not cannot be overloaded. Flytekit follows the same convention used by pandas, numpy, and other libraries.

​
Import

import typing
from flytekit import conditional, task, workflow

​
Simple branch

Choose between calculate_circle_circumference and calculate_circle_area based on whether the input is a fraction (0–1) or a larger value: 
@task
def calculate_circle_circumference(radius: float) -> float:
    return 2 * 3.14159 * radius


@task
def calculate_circle_area(radius: float) -> float:
    return 3.14159 * radius * radius


@workflow
def shape_properties(radius: float = 0.5) -> float:
    return (
        conditional("shape-properties")
        .if_(radius < 1.0)
        .then(calculate_circle_circumference(radius=radius))
        .else_()
        .then(calculate_circle_area(radius=radius))
    )

​
Multiple branches

Use .elif_() to handle multiple distinct conditions. If none matches, the workflow fails: 
@task
def calculate_rectangle_area(length: float, width: float) -> float:
    return length * width


@workflow
def shape_properties_with_multiple_branches(radius: float = 0.5) -> float:
    return (
        conditional("shape-properties-multiple")
        .if_(radius < 0.0)
        .fail("Radius must be non-negative")
        .elif_((radius > 0.0) & (radius < 1.0))
        .then(calculate_circle_circumference(radius=radius))
        .elif_(radius > 1.0)
        .then(calculate_circle_area(radius=radius))
        .else_()
        .fail("Exact zero radius not handled")
    )

​
Consuming the output of a conditional

Pass the result of a conditional to a downstream task just like any other promise: 
@task
def format_result(area: float) -> str:
    return f"The computed area is {area:.4f}"


@workflow
def shape_properties_accept_conditional_output(radius: float = 0.5) -> str:
    result = (
        conditional("shape-properties")
        .if_(radius < 1.0)
        .then(calculate_circle_circumference(radius=radius))
        .else_()
        .then(calculate_circle_area(radius=radius))
    )
    return format_result(area=result)

​
Using a task’s boolean output in a conditional

Use .is_true(), .is_false(), and .is_none() on promise values — unary operations are not supported directly: 
@task
def coin_flip(seed: int) -> bool:
    import random
    random.seed(seed)
    return random.random() > 0.5


@workflow
def boolean_wf(seed: int = 42) -> float:
    result = coin_flip(seed=seed)
    return (
        conditional("coin-flip")
        .if_(result.is_true())
        .then(calculate_circle_circumference(radius=0.5))
        .else_()
        .then(calculate_circle_area(radius=0.5))
    )
Inside a workflow, inputs and outputs are automatically wrapped in a Promise object. This is why result has the .is_true() method — it is a Flytekit Promise, not a plain Python bool.

​
Using boolean workflow inputs

You can pass a boolean directly to a workflow and test it with .is_true(): 
@workflow
def boolean_input_wf(boolean_input: bool = True) -> float:
    return (
        conditional("boolean-input")
        .if_(boolean_input.is_true())
        .then(calculate_circle_circumference(radius=1.0))
        .else_()
        .then(calculate_circle_area(radius=1.0))
    )
The boolean workflow input is also a Promise at workflow-definition time, so it carries the same .is_true() behavior.

​
Nested conditionals

You can nest conditional sections inside the .then() of an outer conditional: 
@workflow
def nested_conditions(radius: float = 0.7) -> float:
    return (
        conditional("outer")
        .if_(radius < 0.0)
        .fail("Negative radius")
        .elif_(radius < 1.0)
        .then(
            conditional("inner")
            .if_(radius < 0.5)
            .then(calculate_circle_circumference(radius=radius))
            .else_()
            .then(calculate_circle_area(radius=radius))
        )
        .else_()
        .then(calculate_circle_area(radius=radius))
    )
Nested conditional sections can only appear inside the .then() part of an outer conditional block.

​
Skipping a branch with a noop

To skip execution in one branch without side effects, use the echo task plugin, which simply passes through its input: 
# Enable the echo plugin in Flyte's config:
# task-plugins:
#   enabled-plugins:
#     - echo

from flytekit.extras.tasks.shell import echo_task

@workflow
def noop_in_conditional(radius: float = 0.4, seed: int = 5) -> typing.Optional[float]:
    flip = coin_flip(seed=seed)
    return (
        conditional("noop-conditional")
        .if_(flip.is_true())
        .then(calculate_circle_circumference(radius=radius))
        .else_()
        .then(echo_task(input_val=None))
    )

​
Run locally

if __name__ == "__main__":
    print(shape_properties(radius=0.5))
    print(shape_properties_with_multiple_branches(radius=11.0))
    print(shape_properties_accept_conditional_output(radius=0.5))
    print(boolean_input_wf(boolean_input=True))
    print(nested_conditions(radius=0.7))

​
Run on the Flyte cluster

pyflyte run --remote \
  https://raw.githubusercontent.com/flyteorg/flytesnacks/656e63d1c8dded3e9e7161c7af6425e9fcd43f56/examples/advanced_composition/advanced_composition/conditional.py \
  shape_properties --radius 3.0

Build docs developers (and LLMs) love

Get started for freeTalk to us