If you have an existing WSGI app, getting this builder to work for you is a piece of 🍰!
Add a vercel.json
file to the root of your application:
{
"builds": [{
"src": "index.py",
"use": "@potatohd/py-vercel",
"config": { "maxLambdaSize": "15mb" }
}]
}
This configuration is doing a few things in the "builds"
part:
"src": "index.py"
This tells Now that there is one entrypoint to build for.index.py
is a file we'll create shortly."use": "@potatohd/py-vercel"
Tell Now to use this builder when deploying your application"config": { "maxLambdaSize": "15mb" }
Bump up the maximum size of the built application to accommodate some larger python WSGI libraries (like Django or Flask). This may not be necessary for you.
Add index.py
to the root of your application. This entrypoint should make
available an object named application
that is an instance of your WSGI
application. E.g.:
# For a Dango app
from django_app.wsgi import application
# Replace `django_app` with the appropriate name to point towards your project's
# wsgi.py file
Look at your framework documentation for help getting access to the WSGI application.
If the WSGI instance isn't named application
you can set the
wsgiApplicationName
configuration option to match your application's name (see
the configuration section below).
That's it, you're ready to go:
$ vercel
> Deploying python-wsgi-app
...
> Success! Deployment ready [57s]
Your project may optionally include a setup.sh
or a post-install.sh
file to declare any
dependencies, e.g.:
# setup.sh
yum install -y gcc mysql-devel
ln -s /usr/lib64/libmariadbclient.a /usr/lib64/libmariadb.a
pip install -t $srcDir $srcDir/django-storages
(setup.sh
runs before python packages installation and post-install.sh
after)
As you can see, there is default variable $srcDir
which contains
path to your project folder. Also, yoy may notice that the server is on
special Amazon Linux 2 runtime. So keep that in mind when you are writing scripts.
Also be aware that on the runtime you must use the -t
argument when installing
python packages.
Your project may optionally include a requirements.txt
file to declare any
dependencies, e.g.:
# requirements.txt
Django >=2.2,<2.3
Be aware that the builder will install Werkzeug
as a requirement of the
handler. This can cause issues if your project requires a different version of
Werkzeug
than the handler.
Select the lambda runtime. Defaults to python3.8
.
{
"builds": [{
"config": { "runtime": "python3.8" }
}]
}
Select the WSGI application to run from your entrypoint. Defaults to
application
.
{
"builds": [{
"config": { "wsgiApplicationName": "application" }
}]
}
Select whether the application is in production mode or not. When set to false, the debugger will appear on error. Defaults to false.
{
"builds": [{
"config": { "production": true }
}]
}
You'll likely want all requests arriving at your deployment url to be routed to your application. You can do this by adding a route rewrite to the Now configuration:
{
"builds": [{
"src": "index.py",
"use": "@potatohd/py-vercel"
}],
"routes" : [{
"src" : "/(.*)", "dest":"/"
}]
}
If having an extra file in your project is troublesome or seems unecessary, it's
also possible to configure Now to use your application directly, without passing
it through index.py
.
If your WSGI application lives in vercel_app/wsgi.py
and is named application
,
then you can configure it as the entrypoint and adjust routes accordingly:
{
"builds": [{
"src": "vercel_app/wsgi.py",
"use": "@potatohd/py-vercel"
}],
"routes" : [{
"src" : "/(.*)", "dest":"/vercel_app/wsgi.py"
}]
}
At the time of writing, Vercel runs on AWS Lambda. This has a number of implications on what libraries will be available to you, notably:
- PostgreSQL, so psycopg2 won't work out of the box
- MySQL, so MySQL adapters won't work out of the box either
- Sqlite, so the built-in Sqlite adapter won't be available
This implementation draws upon work from: