@conda decorator cannot resolve

[kgullikson88]

I am trying to write a Flow using the @conda and @conda_base decorators, and am getting package conflicts that are surprising to me. Minimal example flow:

from metaflow import (
    FlowSpec,
    step,
    conda,
    conda_base,
)

@conda_base(python='3.9.9', libraries={'pandas': '1.3.3'})
class CondaFlow(FlowSpec):
    @conda(libraries={"pyyaml": "6.0"})
    @step
    def start(self):
        self.next(self.end)

    @step
    def end(self):
        pass


if __name__ == "__main__":
    CondaFlow()

My understanding is that this should be effectively equivalent to

conda create -n test python=3.9.9 pandas=1.3.3 pyyaml=6.0

which does resolve. However, when I run the flow using

python conda_flow.py --environment=conda run

I get the following output:

Long Error Message...

Metaflow 2.4.8 executing CondaFlow for user:kgullikson
Validating your flow...
    The graph looks good!
Running pylint...
    Pylint is happy!
Bootstrapping conda environment...(this could take a few minutes)
    Conda ran into an error while setting up environment.:
    Step: start, Error: UnsatisfiableError: The following specifications were found to be incompatible with each other:

    Output in format: Requested package -> Available versions

    Package ca-certificates conflicts for:
    coverage==4.5.4 -> python[version='>=2.7,<2.8.0a0'] -> ca-certificates
    boto3==1.14.47 -> python -> ca-certificates
    python==3.9.9 -> openssl[version='>=3.0.0,<4.0a0'] -> ca-certificates
    click==7.1.2 -> python -> ca-certificates
    requests==2.24.0 -> python -> ca-certificates

    Package pypy3.6 conflicts for:
    requests==2.24.0 -> certifi[version='>=2017.4.17'] -> pypy3.6[version='7.3.0.|7.3.1.|7.3.2.|7.3.3.|>=7.3.1|>=7.3.2|>=7.3.3']
    pandas==1.3.3 -> numpy[version='>=1.19.5,<2.0a0'] -> pypy3.6[version='>=7.3.1|>=7.3.2|>=7.3.3']
    coverage==4.5.4 -> python[version='>=3.6,<3.7.0a0'] -> pypy3.6[version='7.3.0.|7.3.1.|7.3.2.|7.3.3.']
    click==7.1.2 -> python -> pypy3.6[version='7.3.0.|7.3.1.|7.3.2.|7.3.3.']
    boto3==1.14.47 -> python -> pypy3.6[version='7.3.0.|7.3.1.|7.3.2.|7.3.3.|>=7.3.2']

    Package python conflicts for:
    requests==2.24.0 -> certifi[version='>=2017.4.17'] -> python[version='2.7.|3.5.|3.6.|>=2.7,<2.8.0a0|>=3.10,<3.11.0a0|>=3.9,<3.10.0a0|>=3.8,<3.9.0a0|>=3.7,<3.8.0a0|>=3.6,<3.7.0a0|>=3.5,<3.6.0a0|<4.0']
    coverage==4.5.4 -> python[version='>=2.7,<2.8.0a0|>=3.6,<3.7.0a0|>=3.8,<3.9.0a0|>=3.7,<3.8.0a0']
    pyyaml==6.0 -> python[version='>=3.10,<3.11.0a0|>=3.7,<3.8.0a0|>=3.8,<3.9.0a0|>=3.9,<3.10.0a0']
    click==7.1.2 -> python
    python==3.9.9
    boto3==1.14.47 -> jmespath[version='>=0.7.1,<1.0.0'] -> python[version='2.7.|3.5.|3.6.|3.4.|>=3.5,<3.6.0a0|>=2.7,<2.8.0a0|>=3.7,<3.8.0a0|>=3.6,<3.7.0a0|>=3|>=3.9,<3.10.0a0|>=3.8,<3.9.0a0']
    boto3==1.14.47 -> python
    pyyaml==6.0 -> python_abi=3.7[build=_cp37m] -> python[version='3.10.|3.7.|3.8.|3.9.']
    requests==2.24.0 -> python

    Package python_abi conflicts for:
    pyyaml==6.0 -> python_abi[version='3.10.|3.7.|3.8.|3.9.|3.7',build='_cp37m|_cp310|_cp38|_cp39|_pypy37_pp73']
    coverage==4.5.4 -> python[version='>=3.6,<3.7.0a0'] -> python_abi[version='3.6|3.7',build='_pypy37_pp73|_pypy36_pp73']
    click==7.1.2 -> python -> python_abi[version='3.6|3.7',build='_pypy37_pp73|_pypy36_pp73']
    boto3==1.14.47 -> python -> python_abi[version='3.6.|3.6|3.7|3.7.|3.9.|3.8.',build='_cp38|_pypy37_pp73|_pypy36_pp73|_cp37m|_cp39|_cp36m']
    requests==2.24.0 -> certifi[version='>=2017.4.17'] -> python_abi[version='2.7.|3.10.|3.9.|3.8.|3.7.|3.7|3.6.|3.6',build='_pypy37_pp73|_cp39|_cp310|_cp38|_cp37m|_cp36m|_pypy36_pp73|_cp27m']
    pandas==1.3.3 -> numpy[version='>=1.19.5,<2.0a0'] -> python_abi[version='3.10.|3.6.|3.6',build='_cp310|_cp36m|_pypy36_pp73']
    pandas==1.3.3 -> python_abi[version='3.7.|3.7|3.8.|3.9.',build='_cp39|_pypy37_pp73|_cp38|_cp37m']

    Package tzdata conflicts for:
    pandas==1.3.3 -> python[version='>=3.9,<3.10.0a0'] -> tzdata
    pyyaml==6.0 -> python[version='>=3.10,<3.11.0a0'] -> tzdata
    python==3.9.9 -> tzdata
    click==7.1.2 -> python -> tzdata
    requests==2.24.0 -> python -> tzdata
    boto3==1.14.47 -> python -> tzdata

    Package python-dateutil conflicts for:
    boto3==1.14.47 -> botocore[version='>=1.17.47,<1.18.0'] -> python-dateutil[version='>=2.1,<3.0.0']
    pandas==1.3.3 -> python-dateutil[version='>=2.7.3']

    Package certifi conflicts for:
    requests==2.24.0 -> urllib3[version='>=1.21.1,<1.26,!=1.25.0,!=1.25.1'] -> certifi
    requests==2.24.0 -> certifi[version='>=2017.4.17']
    pandas==1.3.3 -> setuptools[version='<60.0.0'] -> certifi[version='>=2016.09|>=2016.9.26']

    Package gdbm conflicts for:
    pyyaml==6.0 -> pypy3.7[version='>=7.3.7'] -> gdbm[version='>=1.18,<1.19.0a0']
    pandas==1.3.3 -> pypy3.7[version='>=7.3.5'] -> gdbm[version='>=1.18,<1.19.0a0']

    Package pypy3.7 conflicts for:
    pyyaml==6.0 -> python[version='>=3.7,<3.8.0a0'] -> pypy3.7[version='7.3.|7.3.3.|7.3.4.|7.3.5.|7.3.7.']
    pyyaml==6.0 -> pypy3.7[version='>=7.3.5|>=7.3.7']

    Package setuptools conflicts for:
    pandas==1.3.3 -> numexpr[version='>=2.7.0'] -> setuptools
    python==3.9.9 -> pip -> setuptools
    pandas==1.3.3 -> setuptools[version='<60.0.0']

    Package urllib3 conflicts for:
    requests==2.24.0 -> urllib3[version='>=1.21.1,<1.26,!=1.25.0,!=1.25.1']
    boto3==1.14.47 -> botocore[version='>=1.17.47,<1.18.0'] -> urllib3[version='>=1.20,<1.26']

[kgullikson88]

So, @jimgoo helped me track it down. It seems the @conda decorator silently adds some additional packages, which makes the above roughly equivalent to

conda create -n test python=3.9.9 pandas=1.3.3 pyyaml=6.0 click=7.1.2 requests=2.24.0 boto3=1.14.47 coverage=4.5.4

That in fact does fail, because there is no coverage=4.5.4 for python=3.9. So really, this issue is that metaflow does not support python>3.8 with the @conda decorator.

More recent versions of coverage do support python 3.9, so this could be fixed by updating the coverage version that is used. For now, I've just downgraded to use python 3.8

[savingoyal]

@kgullikson88 While we address this, you can use Metaflow with Python 3.9 by explicitly adding a dependency on the latest version of coverage=6.2.

[kgullikson88]

Ah, that makes sense!

[npow]

Hi, ran into this issue as well. Where can I see which packages are implicitly added by the @conda decorator?

[kgullikson88]

I saw it in the docs a while ago but can't find now. I'm pretty sure it is just all the metaflow dependencies, which you can find here.