How to deploy multiple Django apps (as subdomains) using nginx and Gunicorn

Question

I have a new 18.04 Droplet running a Django website on nginx. I’d like to use this droplet to serve several websites at subdomains. I have successfully:

Followed this excellent tutorial to set up my first website: https://www.digitalocean.com/community/tutorials/how-to-set-up-django-with-postgres-nginx-and-gunicorn-on-ubuntu-18-04
Configured nginx and DNS to serve (static) websites & subdomains on this server
Set up a second Django app that I can run on port 8000, using Gunicorn as a console process

So where I’m hung up is what I add to the systemd configuration? I tried copy/pasting the [Service] block for my new site but when I try to restart gunicorn I get this error:

systemd[1]: gunicorn.service: Service has more than one ExecStart= setting, which is only allowed for Type=one

Should I create a second *.service file in /etc/systemd/system/? Is there some way to put multiple Django apps into one ExecStart directive? Each app runs in its own virtual environment.

Ben · Accepted Answer

I believe you need to create multiple socket and service files.

I created a separate *.service and *.socket file for each web app
I changed the WorkingDirectory and ExecStart values in the *.service files appropriately

Then I was able to host multiple django projects, each one with a different domain.

From my surface understanding, the process goes something like this. Nginx listening on port 80 receives an HTTP request. This request includes a Host header, which includes myDomain.com. Nginx then looks through it’s sites-available configs for a site that accepts myDomain.com. If it finds one it will look at the appropriate location section in the config file. In my case this includes

proxy_pass http://unix:/run/[MySockFileName].sock;

Some kind of information is then pushed into the unix socket file, named [MySockFileName].sock, inside the /run directory. SystemD notices this and starts up gunicorn if it’s not running already. Gunicorn consumes the information in the socket and after that magically runs python and sends you a rendered web page.

If you find best practices on this for production please let me know.