Hi, we have come across a weird bug with the new version of metaflow while creating a workflow on argo workflows (through mac os). The new version of Metaflow (2.12.30) gets stuck on bootstrapping env and gives a

pip._vendor.resolvelib.resolvers.ResolutionTooDeep: 200000

error whenever we include the mediapipe library inside the package list. So, for example, this is our test code:

from metaflow import FlowSpec, step, project, pypi_base


@pypi_base(
    python="3.11.5",
    packages={
        "mediapipe": "0.10.2",
    },
)
@project(name="test_project")
class FirstFlow(FlowSpec):

    @step
    def start(self):
        print("Inside start function")
        self.next(self.end)

    @step
    def end(self):
        print("Inside end function")


if __name__ == "__main__":
    FirstFlow()

If we run this locally, it works on any version of Metaflow. But if we were to run it on our argo server using the following command:

METAFLOW_PROFILE=gcp PIP_EXTRA_INDEX_URL="<https://pypi.org/simple>" python metaflow_test.py --environment=conda argo-workflows create

it fails with pip resolution error. The surprising thing is that if we downgrade the metaflow version to the old one we were using (2.10.7), the environment gets bootstrapped correctly on argo workflows. So it seems that the new version of metaflow is having problems doing a cross platform bootstrap when the mediapipe library is included in the packages. Our temporary solution is to use the old version to create argo flows if mediapipe package is included in the flow. But we were wondering, what could be the cause of this, and are there any better ways to solve this issue?

@enough-twilight-11147 can you help with which operating system and architecture are you running these commands from?

Hi @square-wire-39606, I am running these commands through my MacBook Pro (Sequoia 15.1.1 (24B91)) with the Apple M1 Pro Chip

@thankful-ambulance-42457 can you look into this?