Hey, I'm trying to implement conditional branching to give the option on the infrastructure used for computation (see code snippet below). I've seen conditional branching is not supported by metaflow, but do you have alternatives for this usecase ? Thank you !

@step
    def evaluate_regression(self) -> None:
        if self.infra:
            self.next(self.evaluate_on_kubernetes)
        else:
            self.next(self.evaluate_locally)

    @kubernetes(
        gpu=1,
        cpu=4,
        memory=16_000,  # 16Gb
        namespace="X",
    )
    @step
    def evaluate_on_kubernetes(self) -> None:
        ...    
        self.next(self.join_target_results)

    @step
    def evaluate_locally(self) -> None:
        ...
        self.next(self.join_target_results)

You can set a var inside

evaluate_regression

and call both steps and have step specific logic based on this infra.

@step
    def evaluate_regression(self) -> None:
        self.infra = True
        self.next(self.evaluate_on_kubernetes, self.evaluate_locally)

    @kubernetes(
        gpu=1,
        cpu=4,
        memory=16_000,  # 16Gb
        namespace="X",
    )
    @step
    def evaluate_on_kubernetes(self) -> None:
        ...
        if self.infra:
            # Infra logic
        self.next(self.join_target_results)


    @step
    def evaluate_locally(self) -> None:
        ...
        if not self.infra:
            # Local logic
        self.next(self.join_target_results)

I agree this is a working solution, but this will still launch a new GKE pod for nothing...

AFAIK sadly metaflow doesn't have a workaround for this

configs may help there too if you know when deploying which one you want to use (ie: if infra is a deploy time value and not a runtime one)