AWS Lambda 'Runtime.ImportModuleError: Unable to import module' / 'No module named' for a Python function (dependencies, layers, wrong architecture). How do I package it correctly?

Cause: The Python runtime could not import your handler module or one of its dependencies. Typical reasons per the Lambda docs: the handler .py file or the dependency folders are not at the root of the .zip (or the layer lacks a top-level python directory); a layer was built with a different Python version than the function's runtime; packages with compiled C/C++ code (NumPy, Pandas and similar) were installed on macOS or Windows, or for the wrong CPU architecture, so the wheel is not compatible with Amazon Linux; or the package is only available as a source distribution and was not built for Lambda. Fix: 1. Zip layout: the handler file and all installed dependencies must sit at the root of the .zip. Example: pip install --target ./package -r requirements.txt, then cd package and zip -r ../deployment.zip . and add your handler with zip deployment.zip lambda_function.py from the project directory. Check the handler setting matches the file and function name (lambda_function.lambda_handler). 2. Compiled packages: install Linux wheels that match the function's architecture. For x86_64: pip install --platform manylinux2014_x86_64 --target=package --implementation cp --python-version 3.x --only-binary=:all: --upgrade PACKAGE. For arm64 use manylinux2014_aarch64. Replace 3.x with your runtime's Python version. If a package has only a source distribution, build it in an Amazon Linux environment, such as an EC2 instance or a container image. 3. Layers: create a python/ directory at the root of the layer zip (pip install -r requirements.txt -t python/), zip it, publish it, and attach the specific layer version ARN. Build the layer with the same Python version as the function, and use Linux-compatible wheels. 4. Check the function's architecture (x86_64 or arm64) matches the wheels you installed. 5. Do not ship __pycache__ folders built elsewhere, and make sure file permissions are readable (644 for files, 755 for directories and executables). 6. Include boto3 and its dependencies in your package if you pin any of them, to avoid version mismatches with the runtime's bundled SDK. Packages in your zip take precedence over layers, which take precedence over runtime-included libraries. 7. Print sys.path in a test invocation if you need to see where Lambda looks for modules. Verify: run a test event; the CloudWatch log shows no ImportModuleError and the handler returns normally. Confirm the unzipped size stays under 250 MB including layers. Sources: https://docs.aws.amazon.com/lambda/latest/dg/python-package.html https://docs.aws.amazon.com/lambda/latest/dg/python-layers.html