Python command-line (CLI) tool to package a Python function for deploying to AWS Lambda, and possibly other cloud platforms.
This tool builds a ZIP file from a virtual environment with all dependencies installed that are to be included in the final deployment asset. If the content is larger than AWS Lambda's maximum unzipped package size of 250 MiB, This tool will then employ the ZIP-inside-ZIP (nested-ZIP) workaround. This allows deploying Lambdas with large dependency packages, especially those with native code compiled extensions like Pandas, PyArrow, etc. The ZIP files are generated reproducibly, ensuring that the same source will always generate a ZIP file with the same hash.
This technique was originally pioneered by serverless-python-requirements, which is a NodeJS (JavaScript) plugin for the Serverless Framework. The technique has been improved here to not require any special imports in your entrypoint source file. That is, no changes are needed to your source code to leverage the nested ZIP deployment.
The motivation for this Python tool is to achieve the same results as serverless-python-requirements but with a purely Python tool. This can simplify and speed up developer and CI/CD workflows.
One important thing that this tool does not do is build the target virtual environment and install all of the dependencies. You must first generate that with a tool like Poetry and the poetry-plugin-bundle.
poetry bundle venv .build/.venv --without dev
package-python-function .build/.venv --output-dir .build/lambdaThe output will be a .zip file named after your project, as described in Output file name.
Use pipx to install:
pipx install package-python-functionpackage-python-function venv_dir [--project PROJECT] [--output-dir OUTPUT_DIR] [--output OUTPUT] [--report REPORT]venv_dir[Required]: The path to the virtual environment to package.--project[Optional]: Path to thepyproject.tomlfile. Omit to use thepyproject.tomlfile in the current working directory.--report[Optional]: Path to write a JSON report file to. Omit to write no report.
--output and --output-dir cannot be used together. If neither is given, the zip is written to the current working
directory.
--output: The full output path of the final zip file.--output-dir: The output directory for the final zip file. The name of the zip file is described in Output file name.
Unless --output gives an exact path, the file written is <output-dir>/<distribution_name>.zip.
distribution_name is the project's name — [project].name, or [tool.poetry].name if that is absent — with each run
of characters outside A-Z a-z 0-9 _ . replaced by a single underscore, following the
PyPA escaping rules.
Case is preserved. A project named My-App produces My_App.zip, not my_app.zip. Note that this differs from the
wheel your build tool produces for the same project, whose filename is lowercased — so a wheel's name is not a safe way
to predict the name of this file.
Pass --report <path> to have the tool write a JSON description of what it produced, so a calling script does not have
to re-derive the output path or re-measure the package.
{
"output_file": "/abs/path/lambda/my_app.zip",
"distribution_name": "my_app",
"output_bytes": 3460000,
"uncompressed_bytes": 412000000,
"compressed_bytes": 3456789,
"nested_zip": false
}| Field | Meaning |
|---|---|
output_file |
Absolute path of the zip that was written. |
distribution_name |
The normalized project name, as described in Output file name. |
output_bytes |
Size of the file at output_file. This is the artifact you deploy. |
uncompressed_bytes |
Total size of the packaged files before compression. This is the figure compared against the AWS Lambda 250 MiB unzipped limit. |
compressed_bytes |
Size of the dependencies zip. Equal to output_bytes unless the nested-zip strategy was used, in which case the outer zip also holds the loader. |
nested_zip |
Whether the nested-zip strategy was used. |
The report is written only when packaging succeeds, so its presence is a reliable signal that the zip is really there.
The ZIP files generated adhere with reproducible builds. This means that file permissions and timestamps are modified inside the ZIP, such that the ZIP will have a deterministic hash. By default, the date is set to 1980-01-01.
Additionally, the tool respects the standardized $SOURCE_DATE_EPOCH environment variable, which will allow you to set that date as needed.
One important caveat is that ZIP files do not support files with timestamps earlier than 1980-01-01 inside them, due to MS-DOS compatibility. Therefore, the tool will throw a SourceDateEpochError is $SOURCE_DATE_EPOCH is below 315532800.
In testing, we found that several file types can leak information from the machine that generated the virtual environment.
To get around this, the tool removes the following files:
**/__pycache/
**/*.dist-info/direct_url.json
**/*.dist-info/RECORD
**/*.pyc
**/*.pyo