Optimizing nginx requests per second

Question

I'm trying to optimize a single core 1GB ram Digital Ocean VPS to handle more requests per second. After some tweaking (workers/gzip etc.) it serves about 15 requests per second. I don't have anything to compare it with but I think this number can be higher.

The stack works like this:

VPS -> Docker container -> nginx (ssl) -> Varnish -> nginx -> uwsgi (Django)

I'm aware of the fact that this is a long chain and that Docker might cause some overhead. However, almost all requests can be handled by Varnish.

These are my tests results:

ab  -kc 100 -n 1000 https://mydomain | grep 'Requests per second'
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests
Requests per second:    18.87 [#/sec] (mean)

I have actually 3 questions:

• Am I correct that 18.87 requests per second is low?
• For a simple Varnished Django blog app, what would be an adequate value (indication)?
• I already applied the recommended tweaks (tuned for my system) from this tutorial. What can I tweak more and how do I figure out the bottlenecks.

Answer 1

First some note about Docker. It is not meant to run a multiple processes in a single docker container. Docker is not a replacement for a VM. It simply allows to run processes in isolation. So the docker diagram should be:

VPS -> docker nginx container -> docker varnish container -> docker django container

To make your life using multiple Docker containers simpler, I would recommend to use Docker-compose. It is not perfect but its an excellent start.

Old but still fundamentally relent blog post about that. Note that some suggestions are no-longer relevant like nsenter since docker exec command is now available but most of the blog post is still correct.

---

As for your performance issues, yes, 18 requests per second is pretty low. However the issue is probably has nothing to do with nginx and is most likely in your Django application and possibly varnish (however very unlikely).

To debug PA issues in Django, I would recommend to use django-debug-toolbar. Most issues in Django are caused by unnecessary SQL queries. You can see them easily in debug toolbar. To solve most them you can use select_related() and prefetch_related. For more detailed analysis, I would also recommend profiling your application. cProfile is a great start. Also some IDEs like PyCharm include built-in profilers so its pretty easy to profile your application to see which functions are taking most of the time which you can optimize. Finally you can use 3rd party tools to profile your application. Even free newrelic account will give you quite a bit of information. Alternatively you can use opbeat which is a new cool kid on the block.