Nginx Caching for Passenger Applications

Suppose you’re running an application on a Passenger + Nginx powered server and you want to add caching.

Perhaps your application has a dynamic, public endpoint but the contents don’t change super-frequently or it isn’t critically-important that the user always gets up-to-the-second accuracy, and you’d like to improve performance with microcaching. How would you do that?

Where you’re at

Diagram showing the Internet connecting to an Nginx+Passenger webserver, connecting to an application written for Ruby, Python, or NodeJS.
Not pictured: the rest of the Internet.

Your configuration might look something like this:

1
2
3
4
5
6
7
server {
  # listen, server_name, ssl, logging etc. directives go here
  # ...

  root               /your/application;
  passenger_enabled  on;
}

What you’re looking for is proxy_cache and its sister directives, but you can’t just insert them here because while Passenger does act act like an upstream proxy (with parallels like e.g. passenger_pass_header which mirrors the behaviour of proxy_pass_header), it doesn’t provide any of the functions you need to implement proxy caching of non-static files.

Where you need to be

Instead, what you need to to is define a second server, mount Passenger in that, and then proxy to that second server. E.g.:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
# Set up a cache
proxy_cache_path /tmp/cache/my-app-cache keys_zone=MyAppCache:10m levels=1:2 inactive=600s max_size=100m;

# Define the actual webserver that listens for Internet traffic:
server {
  # listen, server_name, ssl, logging etc. directives go here
  # ...

  # You can configure different rules by location etc., but here's a simple microcache:
  location / {
    proxy_pass http://127.0.0.1:4863; # Proxy all traffic to the application server defined below
    proxy_cache           MyAppCache; # Use the cache defined above
    proxy_cache_valid     200 3s;     # Treat HTTP 200 responses as valid; cache them for 3 seconds
    proxy_cache_use_stale updating;   # (Optional) send outdated response while background-updating cache
    proxy_cache_lock      on;         # (Optional) only allow one process to update cache at once
  }
}

# (Local-only) application server on an arbitrary port number to act as the upstream proxy:
server {
  listen 127.0.0.1:4863;

  root               /your/application;
  passenger_enabled  on;
}

The two key changes are:

  • Passenger moves to a second server block, localhost-only, on an arbitrary port number (doesn’t need HTTPS, of course, but if your application detects/”expects” HTTPS you might need to tweak your headers).
  • Your main server block proxies to the second as its upstream, and you can add whatever caching directives you like.

Obviously you’ll need to be smarter if you host a mixture of public and private content (e.g. send Vary: headers from your application) and if you want different cache durations on different addresses or types of content, but there are already great guides to help with that. I only wrote this post because I spent some time searching for (nonexistent!) passenger_cache_ etc. rules and wanted to save the next person from the same trouble!

OpenAI-powered Linux shell uses AI to Do What You Mean

This article is a repost promoting content originally published elsewhere. See more things Dan's reposted.

It’s like Alexa/Siri/Cortana for your terminal!

This is a basic Python shell (really, it’s a fancy wrapper over the system shell) that takes a task and asks OpenAI for what Linux bash command to run based on your description. For safety reasons, you can look at the command and cancel before actually running it.

Of all the stupid uses of OpenAI’s GPT-3, this might be the most-amusing. It’s really interesting to see how close – sometimes spot-on – the algorithm comes to writing the right command when you “say what you mean”. Also, how terribly, terribly ill-advised it would be to actually use this for real.

Amusing Line From The Python Manual

I just extracted the following line from the Python documentation (for those of you who don’t know, Python is a relatively-new and somewhat unusual programming language who’s name is derived not from the snake but from Monty Python):

Attempts to pickle unpicklable objects will raise the PicklingError exception; when this happens, an unspecified number of bytes may have already been written to the underlying file.

You have to love any manual with that line in it! It almost beats the famous Fortran line about defining pi as a constant in case it’s value changes.