I\'m setting up pagination for a certain DRF endpoint that works well- however when deployed on my server, which uses HTTPS, the links to the next & previous pages are forme
Since this is the first SO post that comes up in google search result, I think it would be useful to share how I solved it. Mine has a Kubernetes flavor, but under the hood the logic is not so different than the others.
I'm using Kubernetes ingress controller sitting in front of Django, so things might be different to you, but I'll summarize my context so you can see if it's useful to you:
So I got the same issue as OP, the pagination link from DRF is always http
even if I'm hitting the api endpoint using https
. You can see this more clear if you have setup the browser API which comes with the Django REST Framework (DRF), and go to the api root page. I see all links that's generated by DRF are all http
regardless of what protocol I use when visiting the site.
The stack, by order from external world till Django:
settings.py
and start serving requestsnginx.conf
of the ingress controller. In case you don't know how to to this: first figure out the pod name by kuberctl get pods -n
, write that name down, then dump the file by kubectl exec -it -n ingress_controller_namespace ingress_controller_pod_name cat /etc/nginx/nginx.conf > nginx.conf
.nginx.conf
, check if you have proxy_set_header X-Forwarded-Proto $scheme;
in the location
section for your website domain, or similar thing that sets the proxy header X-Forwarded-Proto
properly. By default, the Kubernetes nginx ingress controller should already have this line (or the equivalent) for you.--forwarded-allow-ips="*"
to your gunicorn command, or if you know your nginx server IP, you can limit down to that one, so that gunicorn will forward the headers for you, otherwise gunicorn will strip off those headers. So you end up with command like gunicorn django_server.wsgi:application --forwarded-allow-ips="*" --workers=${PROPER_WORKER_NUM} --log-level info --bind 0.0.0.0:8001
. You can have 4 workers by specifying workers=4
, usually depends on the CPU you have on the server. The --log-level info
is just for debug purpose, it's optional.SECURE_PROXY_SSL_HEADER = ('HTTP_X_FORWARDED_PROTO', 'https')
in settings.py
, which other answer has mentioned. You may wonder why the prefix HTTP_
while in nginx our header name is X-Forwarded-Proto
. This is because WSGI will add the prefix to headers that it recognizes. This line basically says, if the header HTTP_X_FORWARDED_PROTO
equals to the string "https"
, then Django recognizes the request as secure. This will affect several behaviors in Django, e.g. you will get request.is_secure == True
, request.build_absolute_uri(None) == 'https://...'
, and most important, Django REST Framework pagination link will now use https
! (As long as you indeed hit the api by https
)Alright, now you can test again. If DRF gives you https now - congrats. Or, if you are same as me, after trying above, still no luck - DRF still generating the damn http
links. I want to share some tipbits when I was debugging like hell:
Print out the following values in Django, either by print()
, if you have DEBUG=True
and have access to the server log, or just pass them to template context and show them in a html page, if that makes it easier for you to test in a production environment where SSL is enabled.
request.is_secure()
: if you're failing to get DRF to use http
, chances are you are getting False
.request.META
gives you all headers that Django received. Did the header HTTP_X_FORWARDED_PROTO
show up? What value is it?
HTTP_X_FORWARDED_PROTO
is https,https
, which tells me that it somehow having duplicated values, and I probably have proxy_set_header
set twice, and Bingo! By default K8 ingress controller already has this line, and since I added it again through a location-snippet
, now I have two lines of proxy_set_header X-Forwarded-Proto $scheme;
in total. After removing my snippet, DRF shows https
and I was so so happy to get this done. It wasn't that straight forward because I thought proxy_set_header
will overwrite and "set" the header value. But seems like it just keep appending.