Serverless Python Requirements

A Serverless v1.x plugin to automatically bundle dependencies from requirements.txt and make them available in your PYTHONPATH.

---

Originally developed by Capital One, now maintained in scope of Serverless, Inc

Capital One considers itself the bank a technology company would build. It's delivering best-in-class innovation so that its millions of customers can manage their finances with ease. Capital One is all-in on the cloud and is a leader in the adoption of open source, RESTful APIs, microservices and containers. We build our own products and release them with a speed and agility that allows us to get new customer experiences to market quickly. Our engineers use artificial intelligence and machine learning to transform real-time data, software and algorithms into the future of finance, reimagined.

---

Install

sls plugin install -n serverless-python-requirements

This will automatically add the plugin to your project's package.json and the plugins section of its serverless.yml. That's all that's needed for basic use! The plugin will now bundle your python dependencies specified in your requirements.txt or Pipfile when you run sls deploy.

For a more in depth introduction on how to use this plugin, check out this post on the Serverless Blog

If you're on a mac, check out these notes about using python installed by brew.

Cross compiling

Compiling non-pure-Python modules or fetching their manylinux wheels is supported on non-linux OSs via the use of Docker and the docker-lambda image. To enable docker usage, add the following to your serverless.yml:

custom:
  pythonRequirements:
    dockerizePip: true

The dockerizePip option supports a special case in addition to booleans of 'non-linux' which makes it dockerize only on non-linux environments.

To utilize your own Docker container instead of the default, add the following to your serverless.yml:

custom:
  pythonRequirements:
    dockerImage: <image name>:tag

This must be the full image name and tag to use, including the runtime specific tag if applicable.

Alternatively, you can define your Docker image in your own Dockerfile and add the following to your serverless.yml:

custom:
  pythonRequirements:
    dockerFile: ./path/to/Dockerfile

With Dockerfile the path to the Dockerfile that must be in the current folder (or a subfolder). Please note the dockerImage and the dockerFile are mutually exclusive.

To install requirements from private git repositories, add the following to your serverless.yml:

custom:
  pythonRequirements:
    dockerizePip: true
    dockerSsh: true

The dockerSsh option will mount your $HOME/.ssh/id_rsa and $HOME/.ssh/known_hosts as a volume in the docker container. If your SSH key is password protected, you can use ssh-agent because $SSH_AUTH_SOCK is also mounted & the env var set. It is important that the host of your private repositories has already been added in your $HOME/.ssh/known_hosts file, as the install process will fail otherwise due to host authenticity failure.

You can also pass environment variables to docker by specifying them in dockerEnv option:

custom:
  pythonRequirements:
    dockerEnv:
      - https_proxy

🏁 Windows notes

✨🍰✨ Pipenv support

If you include a Pipfile and have pipenv installed instead of a requirements.txt this will use pipenv lock -r to generate them. It is fully compatible with all options such as zip and dockerizePip. If you don't want this plugin to generate it for you, set the following option:

custom:
  pythonRequirements:
    usePipenv: false

✨📝✨ Poetry support

If you include a pyproject.toml and have poetry installed instead of a requirements.txt this will use poetry export --without-hashes -f requirements.txt -o requirements.txt --with-credentials to generate them. It is fully compatible with all options such as zip and dockerizePip. If you don't want this plugin to generate it for you, set the following option:

custom:
  pythonRequirements:
    usePoetry: false

Poetry with git dependencies

Poetry by default generates the exported requirements.txt file with -e and that breaks pip with -t parameter (used to install all requirements in a specific folder). In order to fix that we remove all -e from the generated file but, for that to work you need to add the git dependencies in a specific way.

Instead of:

[tool.poetry.dependencies]
bottle = {git = "[email protected]/bottlepy/bottle.git", tag = "0.12.16"}

Use:

[tool.poetry.dependencies]
bottle = {git = "https://[email protected]/bottlepy/bottle.git", tag = "0.12.16"}

Or, if you have an SSH key configured:

[tool.poetry.dependencies]
bottle = {git = "ssh://[email protected]/bottlepy/bottle.git", tag = "0.12.16"}

Dealing with Lambda's size limitations

To help deal with potentially large dependencies (for example: numpy, scipy and scikit-learn) there is support for compressing the libraries. This does require a minor change to your code to decompress them. To enable this add the following to your serverless.yml:

custom:
  pythonRequirements:
    zip: true

and add this to your handler module before any code that imports your deps:

try:
  import unzip_requirements
except ImportError:
  pass

Slim Package

Works on non 'win32' environments: Docker, WSL are included To remove the tests, information and caches from the installed packages, enable the slim option. This will: strip the .so files, remove __pycache__ and dist-info directories as well as .pyc and .pyo files.

custom:
  pythonRequirements:
    slim: true

Custom Removal Patterns

To specify additional directories to remove from the installed packages, define a list of patterns in the serverless config using the slimPatterns option and glob syntax. These patterns will be added to the default ones (**/*.py[c|o], **/__pycache__*, **/*.dist-info*). Note, the glob syntax matches against whole paths, so to match a file in any directory, start your pattern with **/.

custom:
  pythonRequirements:
    slim: true
    slimPatterns:
      - '**/*.egg-info*'

To overwrite the default patterns set the option slimPatternsAppendDefaults to false (true by default).

custom:
  pythonRequirements:
    slim: true
    slimPatternsAppendDefaults: false
    slimPatterns:
      - '**/*.egg-info*'

This will remove all folders within the installed requirements that match the names in slimPatterns

Option not to strip binaries

In some cases, stripping binaries leads to problems like "ELF load command address/offset not properly aligned", even when done in the Docker environment. You can still slim down the package without *.so files with

custom:
  pythonRequirements:
    slim: true
    strip: false

Lambda Layer

Another method for dealing with large dependencies is to put them into a Lambda Layer. Simply add the layer option to the configuration.

custom:
  pythonRequirements:
    layer: true

The requirements will be zipped up and a layer will be created automatically. Now just add the reference to the functions that will use the layer.

functions:
  hello:
    handler: handler.hello
    layers:
      - Ref: PythonRequirementsLambdaLayer

If the layer requires additional or custom configuration, add them onto the layer option.

custom:
  pythonRequirements:
    layer:
      name: ${self:provider.stage}-layerName
      description: Python requirements lambda layer
      compatibleRuntimes:
        - python3.7
      licenseInfo: GPLv3
      allowedAccounts:
        - '*'

Omitting Packages

You can omit a package from deployment with the noDeploy option. Note that dependencies of omitted packages must explicitly be omitted too.

This example makes it instead omit pytest:

custom:
  pythonRequirements:
    noDeploy:
      - pytest

Extra Config Options

Caching

You can enable two kinds of caching with this plugin which are currently both ENABLED by default. First, a download cache that will cache downloads that pip needs to compile the packages. And second, a what we call "static caching" which caches output of pip after compiling everything for your requirements file. Since generally requirements.txt files rarely change, you will often see large amounts of speed improvements when enabling the static cache feature. These caches will be shared between all your projects if no custom cacheLocation is specified (see below).

Please note: This has replaced the previously recommended usage of "--cache-dir" in the pipCmdExtraArgs

custom:
  pythonRequirements:
    useDownloadCache: true
    useStaticCache: true

Other caching options

There are two additional options related to caching. You can specify where in your system that this plugin caches with the cacheLocation option. By default it will figure out automatically where based on your username and your OS to store the cache via the appdirectory module. Additionally, you can specify how many max static caches to store with staticCacheMaxVersions, as a simple attempt to limit disk space usage for caching. This is DISABLED (set to 0) by default. Example:

custom:
  pythonRequirements:
    useStaticCache: true
    useDownloadCache: true
    cacheLocation: '/home/user/.my_cache_goes_here'
    staticCacheMaxVersions: 10

Extra pip arguments

You can specify extra arguments supported by pip to be passed to pip like this:

custom:
  pythonRequirements:
    pipCmdExtraArgs:
      - --compile

Extra Docker arguments

You can specify extra arguments to be passed to docker build during the build step, and docker run during the dockerized pip install step:

custom:
  pythonRequirements:
    dockerizePip: true
    dockerBuildCmdExtraArgs: ['--build-arg', 'MY_GREAT_ARG=123']
    dockerRunCmdExtraArgs: ['-v', '${env:PWD}:/my-app']

Customize requirements file name

Some pip workflows involve using requirements files not named requirements.txt. To support these, this plugin has the following option:

custom:
  pythonRequirements:
    fileName: requirements-prod.txt

Per-function requirements

If you have different python functions, with different sets of requirements, you can avoid including all the unecessary dependencies of your functions by using the following structure:

├── serverless.yml
├── function1
│      ├── requirements.txt
│      └── index.py
└── function2
       ├── requirements.txt
       └── index.py

With the content of your serverless.yml containing:

package:
  individually: true

functions:
  func1:
    handler: index.handler
    module: function1
  func2:
    handler: index.handler
    module: function2

The result is 2 zip archives, with only the requirements for function1 in the first one, and only the requirements for function2 in the second one.

Quick notes on the config file:

• The module field must be used to tell the plugin where to find the requirements.txt file for each function.
• The handler field must not be prefixed by the folder name (already known through module) as the root of the zip artifact is already the path to your function.

Customize Python executable

Sometimes your Python executable isn't available on your $PATH as python2.7 or python3.6 (for example, windows or using pyenv). To support this, this plugin has the following option:

custom:
  pythonRequirements:
    pythonBin: /opt/python3.6/bin/python

Vendor library directory

For certain libraries, default packaging produces too large an installation, even when zipping. In those cases it may be necessary to tailor make a version of the module. In that case you can store them in a directory and use the vendor option, and the plugin will copy them along with all the other dependencies to install:

custom:
  pythonRequirements:
    vendor: ./vendored-libraries
functions:
  hello:
    handler: hello.handler
    vendor: ./hello-vendor # The option is also available at the function level

Manual invocations

The .requirements and requirements.zip(if using zip support) files are left behind to speed things up on subsequent deploys. To clean them up, run sls requirements clean. You can also create them (and unzip_requirements if using zip support) manually with sls requirements install.

Invalidate requirements caches on package

If you are using your own Python library, you have to cleanup .requirements on any update. You can use the following option to cleanup .requirements everytime you package.

custom:
  pythonRequirements:
    invalidateCaches: true

🍎🍺🐍 Mac Brew installed Python notes

Brew wilfully breaks the --target option with no seeming intention to fix it which causes issues since this uses that option. There are a few easy workarounds for this:

• Install Python from python.org and specify it with the pythonBin option.

OR

• Create a virtualenv and activate it while using serverless.

OR

• Install Docker and use the dockerizePip option.

Also, brew seems to cause issues with pipenv, so make sure you install pipenv using pip.

🏁 Windows dockerizePip notes

For usage of dockerizePip on Windows do Step 1 only if running serverless on windows, or do both Step 1 & 2 if running serverless inside WSL.

1. Enabling shared volume in Windows Docker Taskbar settings
2. Installing the Docker client on Windows Subsystem for Linux (Ubuntu)

Native Code Dependencies During Build

Some Python packages require extra OS dependencies to build successfully. To deal with this, replace the default image (lambci/lambda:python3.6) with a Dockerfile like:

FROM lambci/lambda:build-python3.6

# Install your dependencies
RUN yum -y install mysql-devel

Then update your serverless.yml:

custom:
  pythonRequirements:
    dockerFile: Dockerfile

Native Code Dependencies During Runtime

Some Python packages require extra OS libraries (*.so files) at runtime. You need to manually include these files in the root directory of your Serverless package. The simplest way to do this is to use the dockerExtraFiles option.

For instance, the mysqlclient package requires libmysqlclient.so.1020. If you use the Dockerfile from the previous section, add an item to the dockerExtraFiles option in your serverless.yml:

custom:
  pythonRequirements:
    dockerExtraFiles:
      - /usr/lib64/mysql57/libmysqlclient.so.1020

Then verify the library gets included in your package:

sls package
zipinfo .serverless/xxx.zip

If you can't see the library, you might need to adjust your package include/exclude configuration in serverless.yml.

Optimising packaging time

If you wish to exclude most of the files in your project, and only include the source files of your lambdas and their dependencies you may well use an approach like this:

package:
  individually: false
  include:
    - './src/lambda_one/**'
    - './src/lambda_two/**'
  exclude:
    - '**'

This will be very slow. Serverless adds a default "**" include. If you are using the cacheLocation parameter to this plugin, this will result in all of the cached files' names being loaded and then subsequently discarded because of the exclude pattern. To avoid this happening you can add a negated include pattern, as is observed in serverless/serverless#5825.

Use this approach instead:

package:
  individually: false
  include:
    - '!./**'
    - './src/lambda_one/**'
    - './src/lambda_two/**'
  exclude:
    - '**'

Contributors

• @dschep - Lead developer & original maintainer
• @azurelogic - logging & documentation fixes
• @abetomo - style & linting
• @angstwad - deploy --function support
• @mather - the cache invalidation option
• @rmax - the extra pip args option
• @bsamuel-ui - Python 3 support, current maintainer
• @suxor42 - fixing permission issues with Docker on Linux
• @mbeltran213 - fixing docker linux -u option bug
• @Tethik - adding usePipenv option
• @miketheman - fixing bug with includes when using zip option, update eslint,
• @wattdave - fixing bug when using deploymentBucket
• @heri16 - fixing Docker support in Windows
• @ryansb - package individually support
• @cgrimal - Private SSH Repo access in Docker, dockerFile option to build a custom docker image, real per-function requirements, and the vendor option
• @kichik - Imposed windows & noDeploy support, switched to adding files straight to zip instead of creating symlinks, and improved pip cache support when using docker.
• @dee-me-tree-or-love - the slim package option
• @alexjurkiewicz - docs about docker workflows
• @andrewfarley - Implemented download caching and static caching
• @bweigel - adding the slimPatternsAppendDefaults option & fixing per-function packaging when some functions don't have requirements & Porting tests from bats to js!
• Poetry support
  • @squaresurf
  • @drice
  • @ofercaspi
  • @tpansino
• @david-mk-lawrence - added Lambda Layer support
• @bryantbriggs - Fixing CI/CD
• @jacksgt - Fixing pip issues
• @lephuongbg - Fixing single function deployment
• @rileypriddle - Introducing schema validation for module property