App Packaging¶

App Directories¶

You have two options to structure application specific code/config:

• app.py - This file includes all your route information and is always included in the deployment package.

• chalicelib/ - This directory (if it exists) is included in the deployment package. This is where you can add config files and additional application modules if you prefer not to have all your app code in the app.py file.

See Multifile Support for more info on the chalicelib/ directory. Both the app.py and the chalicelib/ directory are intended for code that you write yourself.

3rd Party Packages¶

When handling third party packages, you can have Chalice manage these files as part of the deployment package of your Lambda function, or as a separate Lambda layer that’s shared by all your Lambda functions. See the Examples section for examples of how this works.

There are two options for handling python package dependencies:

• requirements.txt - During the packaging process, Chalice will install any packages it finds or can build compatible wheels for. Specifically all pure python packages as well as all packages that upload wheel files for the manylinux1_x86_64 platform will be automatically installable.

• vendor/ - The contents of this directory are automatically added to your deployment package and its location in your Lambda functions will depend on whether you are using automatic Lambda Layers, described in the Automatic Lambda Layers section.

Chalice will also check for an optional vendor/ directory in the project root directory. The contents of this directory are automatically included in the top level of the deployment package (see Examples for specific examples). The vendor/ directory is helpful in these scenarios:

• You need to include custom packages or binary content that is not accessible via pip. These may be internal packages that aren’t public.

• Wheel files are not available for a package you need from pip.

• A package is installable with requirements.txt but has optional c extensions. Chalice can build the dependency without the c extensions, but if you want better performance you can vendor a version that is compiled.

As a general rule of thumb, code that you write goes in either app.py or chalicelib/, and dependencies are either specified in requirements.txt or placed in the vendor/ directory.

.
├── app.py
└── vendor
    └── myimage.png

@app.lambda_function()
def handler(event, context):
    with open('myimage.png') as f:
        do_something(f)

'/opt/python/lib/python%s.%s/site-packages' % sys.version_info[:2]

@app.lambda_function()
def handler(event, context):
    with open_vendor_file('myimage.png') as f:
        do_something(f)

def open_vendor_file(filename):
    directories = [
        '.',
        '/opt/python/lib/python%s.%s/site-packages' % sys.version_info[:2]
    ]
    for dirname in directories:
        full_path = os.path.join(dirname, filename)
        if os.path.isfile(full_path):
            return open(full_path)

Environment Variables¶

As part of the packaging and deployment process, Chalice will import your app.py file. This will result in any top level module code being executed. This can sometimes have undesireable behavior. When running any Chalice CLI commands, a AWS_CHALICE_CLI_MODE environment variable is set. You can check if this env var is set in your app.py if you have code that you don’t want to run whenever your app is packaged and deployed.

import os

app = Chalice(app_name='testimport')

expensive_connection = None
if 'AWS_CHALICE_CLI_MODE' not in os.environ:
    # We're running in Lambda, we want to start up
    # our connection to our DB.
    expensive_connection = ConnectToDB()

Chalice will also set any environment variables specified in your global or stage specific configuration whenever your app is packaged and deployed. Per-Lambda function environment variables are not set when importing your app (this would require importing your application for each Lambda function). For example, given the config below you would be able to access the STAGE_VAR environment variable but not the PER_FUNCTION variable during the building/packaging process when Chalice imports your application. This can be useful if you want to move configuration or resource names out of your app.py file.

{
  "stages": {
    "dev": {
      "environment_variables": {
        "STAGE_VAR": "stage-var"
      }
      "api_gateway_stage": "api",
      "lambda_functions": {
        "foo": {
          "environment_variables": {"PER_FUNCTION": "per-function"}
        }
      }
    }
  },
  "version": "2.0",
  "app_name": "demo"
}

This only applies to the packaging stage. When the foo function is invoked on Lambda, the PER_FUNCTION environment variable will be set as expected.

Examples¶

Suppose I have the following app structure:

.
├── app.py
├── chalicelib
│   ├── __init__.py
│   └── utils.py
├── requirements.txt
└── vendor
    ├── myimage.png
    └── internalpackage
        └── __init__.py

And the requirements.txt file had one requirement:

$ cat requirements.txt
sortedcontainers==1.5.4

With the default behavior of not using layers ("automatic_layer": false), the final deployment package directory structure would look like this:

deployment.zip
.
├── app.py
├── chalicelib
│   ├── __init__.py
│   └── utils.py
├── myimage.png
├── internalpackage
│   └── __init__.py
└── sortedcontainers
    └── __init__.py

This directory structure is then zipped up and sent to AWS Lambda during the deployment process. Suppose our application had two Lambda functions. Each Lambda function has its own copy of the application deployment package, as shown in the architecture diagram below.

If you are using "automatic_layer": true, then two zip files will be created. The deployment package used for the Lambda function will be:

deployment.zip
.
├── app.py
└── chalicelib
    ├── __init__.py
    └── utils.py

And the zip file for the shared lambda layer will look like this:

layer-deployment.zip
.
└── python
    └── lib
        └── python3.7
            └── site-packages
                ├── myimage.png
                ├── internalpackage
                │   └── __init__.py
                └── sortedcontainers
                    └── __init__.py

Below is an updated diagram of the same Chalice application using automatic layers. Both functions now share the same Lambda layer that contains the third party packages used by the application.

Cryptography Example¶

Below shows an example of how to use the cryptography 1.9 package in a Chalice app for the python3.6 lambda environment.

Suppose you are on a Mac or Windows and want to deploy a Chalice app that depends on the cryptography==1.9 package. If you simply add it to your requirements.txt file and try to deploy it with chalice deploy you will get the following warning during deployment:

$ cat requirements.txt
cryptography==1.9
$ chalice deploy
Updating IAM policy.
Updating lambda function...
Creating deployment package.

Could not install dependencies:
cryptography==1.9
You will have to build these yourself and vendor them in
the chalice vendor folder.

Your deployment will continue but may not work correctly
if missing dependencies are not present. For more information:
http://aws.github.io/chalice/topics/packaging.html

This happened because the cryptography version 1.9 does not have wheel files available on PyPi, and has C extensions. Since we are not on the same platform as AWS Lambda, the compiled C extensions Chalice built were not compatible. To get around this we are going to leverage the vendor/ directory, and build the cryptography package on a compatible linux system.

You can do this yourself by building cryptography on an Amazon Linux instance running in EC2. All of the following commands were run inside a python 3.6 virtual environment.

• Download the source first:

  $ pip download cryptography==1.9
  

  This will download all the requirements into the current working directory. The directory should have the following contents:

  • asn1crypto-0.22.0-py2.py3-none-any.whl

  • cffi-1.10.0-cp36-cp36m-manylinux1_x86_64.whl

  • cryptography-1.9.tar.gz

  • idna-2.5-py2.py3-none-any.whl

  • pycparser-2.17.tar.gz

  • six-1.10.0-py2.py3-none-any.whl

  This is a complete set of dependencies required for the cryptography package. Most of these packages have wheels that were downloaded, which means they can simply be put in the requirements.txt and Chalice will take care of downloading them. That leaves cryptography itself and pycparser as the only two that did not have a wheel file available for download.

• Next build the cryptography source package into a wheel file:

  $ pip wheel cryptography-1.9.tar.gz
  

  This will take a few seconds and build a wheel file for both cryptography and pycparser. The directory should now have two additional wheel files:

  • cryptography-1.9-cp36-cp36m-linux_x86_64.whl

  • pycparser-2.17-py2.py3-none-any.whl

  The cryptography wheel file has been built with a compatible architecture for Lambda (linux_x86_64) and the pycparser has been built for any architecture which means it can also be automatically packaged by Chalice if it is listed in the requirements.txt file.

• Download the cryptography wheel file from the Amazon Linux instance and unzip it into the vendor/ directory in the root directory of your Chalice app.

  You should now have a project directory that looks like this:

  $ tree
  .
  ├── app.py
  ├── requirements.txt
  └── vendor
      ├── cryptography
      │   ├── ... Lots of files
      │
      └── cryptography-1.9.dist-info
          ├── DESCRIPTION.rst
          ├── METADATA
          ├── RECORD
          ├── WHEEL
          ├── entry_points.txt
          ├── metadata.json
          └── top_level.txt
  

  The requirements.txt file should look like this:

  $ cat requirements.txt
  cffi==1.10.0
  six==1.10.0
  asn1crypto==0.22.0
  idna==2.5
  pycparser==2.17
  

  In your app.py file you can now import cryptography, and these dependencies will all get included when the chalice deploy command is run.