$ npm install --save serverless-package-python-functions
# serverless.yml
plugins:
- serverless-package-python-functions
A Serverless Framework plugin for packaging Python Lambda functions with only the dependencies they need.
This plugin makes it easy to manage function-level and service-level dependencies for your awesome Python Serverless project
Let's consider the following project structure
your-awesome-project/
├── common_files
│ ├── common1.py
│ └── common2.py
├── function1
│ ├── lambda.py
│ └── requirements.txt # with simplejson library
├── function2
│ ├── lambda.py
│ └── requirements.txt
├── requirements.txt # with requests library
└── serverless.yml
This project has:
- two functions,
function1
andfunction2
, each with their ownrequirements.txt
files. function1's requirements.txt lists the simplejson pip package - Code common to both
function1
andfunction2
in a directory namedcommon_files
- A top-level
requirements.txt
file with pip dependencies common to both functions, e.g requests library
This plugin will package your functions into individual zip files that look like:
├── lambda.py # function-level code
├── requirements.txt
├── common1.py # service-level code
├── common2.py
├── simplejson # function-level dependencies
├── simplejson-3.10.0.dist-info
├── requests # service-level dependencies
└── requests-2.13.0.dist-info
So that the below code
import common1, common2, requests, simplejson
in function1/lambda.py works like works like a charm!
The plugin also supports packaging your dependencies using a Docker Image that replicates your cloud providers environment, allowing you easily work with platform-dependent libraries like numpy.
The plugin handles the creation of the artifact zip files for your Serverless functions.
When serverless deploy
is run, the plugin will:
- create a build directory for each function
- copy the appropriate function-level and service-level code you specify into each function's build directory
- Download the appropriate function-level and service-level pip dependencies into each function's build directory
- Create zip files of each functions build directory
The Serverless framework will then pickup each zip file and upload it to your provider.
Here's a simple serverless.yml
configuration for this plugin, assuming the project structure above
one of the functions we add -${opt:stage}
to the name in order to append the stage to the function name
service: your-awesome-project
package:
individually: true
plugins:
- serverless-package-python-functions
custom:
pkgPyFuncs: # plugin configuration
buildDir: _build
requirementsFile: 'requirements.txt'
globalRequirements:
- ./requirements.txt
globalIncludes:
- ./common_files
cleanup: true
functions:
function1:
name: function1-${opt:stage}
handler: lambda.handler
package:
include:
- function1
artifact: ${self:custom.pkgPyFuncs.buildDir}/function1.zip
function2:
name: function2
handler: lambda.handler
package:
include:
- function2
artifact: ${self:custom.pkgPyFuncs.buildDir}/function2.zip
The plugin configurations are simple:
Configuration | Description | Optional? |
---|---|---|
buildDir | Path to a build directory relative to project root, e.g. build | No |
requirementsFile | The name of the requirements file used for function-level requirements. All function-level requirements files must use the name specified here. | Yes. Defaults to requirements.txt |
globalRequirements | A list of paths to files containing service-level pip requirements. | Yes |
globalIncludes | A list of paths to folders containing service-level code files (i.e. code common to all functions). Only the folders contents will be packaged, not the folder itself. Paths to files are not currently supported. | Yes |
useDocker | Boolean indicating whether to package pip dependencies using Docker. Set this to true if your project uses platform-specific compiled libraries like numpy. Requires a Docker installation. | Yes. Defaults to false |
dockerImage | The Docker image to use to compile functions if useDocker is set to true . Must be specified as repository:tag . If the image doesn't exist on the system, it will be downloaded. The initial download may take some time. |
Yes. Defaults to lambci/lambda:build-${provider.runtime} |
containerName | The desired name for the Docker container. | Yes. Defaults to serverless-package-python-functions |
At the function level, you:
- Specify
name
to give your function a name. The plugin uses the function's name as the name of the zip artifact - Use
include
to specify what function-level files you want to include in your artifact. Simply specifying the path to the function's folder will include every file in the folder in the function's zip artifact - Use
artifact
to tell Serverless where to find the zip artifact. The plugin creates the zip artifact for the function atbuildDir
/name
.zip, so using${self:custom.pkgPyFuncs.buildDir}/[function-name-here].zip
is advised.
At the package level, you may need to:
- Specify the
individually
parameter astrue
to ensure that zip artifacts are generated properly. You may need this if you are getting file not found errors about your zip artifact.
Now, you may be wondering, doesn't the Serverless documentation say:
Serverless won't zip your service if [artifact] is configured and therefore exclude and include will be ignored. Either you use artifact or include / exclude.
Yes, that is correct and is actually awesome! Since Serverless ignores include
/exclude
silently when artifact
is specified, it allows this plugin take advantage of the include
property to provide you with a familiar interface for specifying function-level dependencies. So while this plugin uses include
to determine what goes in your artifact, all Serverless cares about is the artifact that this plugin creates when it executes.
The last thing that your keen eye may have noticed from the example serverless.yml
above is that handler
is specified simply as lambda.handler
not ${self:custom.pkgPyFuncs.buildDir}/function/lambda.hadler
or function/lambda.handler
. This is because the plugin zips your artifacts such that /path/to/function is the root of the zip file. Combined with the fact that it uses pip install -t
to download pip dependencies directly to the top level of the zip file, this makes imports significantly simpler for your project.
Furthermore, since pip install -t
downloads the actual pip package files into a folder, this plugin works without the need for virtualenv