Hey all, Im trying to use Pants to build an AWS Lambda function. Ive been able to pretty much get it all working, except for one catch. Our code relies transiently on the

cryptography

library, and as I understand it, AWS Lambdas require a specific binary for that (and probably other) libraries. I was able to resolve that pre-Pants by following the instructions on this Github issue, but Im struggling to figure out how to get that working with the

python_awslambda

target in Pants. Pre-Pants I can run this pip command to fetch the correct binary

pip install --platform manylinux2014_x86_64 --implementation cp --python 3.9 --only-binary=:all: --upgrade --target ./ cryptography==38.0.1

. But I dont understand how to translate this information to Pants and Pex. Im doing my best to understand, but pretty much everything Im doing is a stab in the dark 😂 I assume I need to do something with the

complete_platforms

argument in the

python_awslambda

target. So I created a

file

target to a

lambda_platform.json

file, and provided that to the

complete_platform

argument. The json looks like this:

{
  "marker_environment": {
    "platform_release": "manylinux2014_x86_64",
    "implementation_name": "cp",
    "python_version": "3.9"
  },
  "compatible_tags": [
    "cp39-abi3-manylinux2014_x86_64"
  ]
}

The package goal prints the correct platform file, however if I look at the cryptography dist info in the generated lambda, the

WHEEL

file differs from what is generated when I run the pip command, and it doesnt seem to include the

manylinux2014_x86_64

platform. So it doesnt appear to have worked. Does anyone have any advice or thoughts here? Am I on the right track? (Side note, changing the lambda_platform.json contents doesnt trigger a new package. I would've thought Pants would re-build if that dependency changed)

I don't know if it's the cause of the issue you're seeing here, but we package lambdas (for Python 3.9) that include

cryptography

, and it works fine for us on macOS and Linux. We use a complete platform file that's generated by running the appropriate pex command within the lambda environment itself, which ends up including many more elements in

compatible_tags

, presumably all tags that are compatible (as well as different

marker_environment

values). Specifically, we set up a lambda containing the following, invoked it manually, and then copying the output into the

.json

file:

import subprocess

def lambda_handler(event, context):
    subprocess.run(
        """
        pip install --target=/tmp/subdir pex
        PYTHONPATH=/tmp/subdir /tmp/subdir/bin/pex3 interpreter inspect --markers --tags
        """,
        shell=True
    )
    return {
        'statusCode': 200,
        'body': "{}",
    }

Untitled.txt

I wonder if you’re hitting issue #17621. I vaguely recall seeing similar errors before adding a workaround for that (see the comment at the end of the issue for one possibility). Are you running from a Linux machine with cpython 3.9 installed?

Ah! That worked!! Thanks so much! I have no idea how long it would've taken to figure that out on my own

Woohoo! No worries I'm glad the hours of confusion I went through saved someone else that experience 😅

This thread saved me a lot of work and improved my understanding of wheel compatibility. Thank you so much ❤️