Reverse Proxy

<slideshow style="nobleprog" headingmark="⌘" incmark="…" scaled="true" font="Trebuchet MS" >

title

Reverse Proxy

author

Bernard Szlachta (NobleProg Ltd)

</slideshow>

nginx as a reverse proxy ⌘

• A popular use of nginx
• Minimises concurrent connections to proxy backend
• Put nginx in front of Tomcat

• nginx serves all static content (images, css, javascript)
• nginx proxies to Tomcat for dynamic content

A simple example ⌘

location /static/ {
   root /var/www/static/;
 }

location / {
   proxy_pass http://localhost:8000/;
 }

Automatic URI rewriting ⌘

• On the previous example the path would be unchanged because it is at the root
• If a path is specified in location and target, by default it is rewritten

 location /foo/ {
   proxy_pass http://localhost:8000/bar/;
 }

• The proxy target will see the request as though it had been made for /bar/

Rewriting Server Responses ⌘

• Sometimes applications will generate 301/302
• Often this will include a path calculated from the local settings
• When used behind a proxy this may be wrong

Server response problem example ⌘

location /foo/ {
  proxy_pass http://localhost:8000/bar/;
}

$ curl -I http://localhost/foo/
HTTP/1.1 302 Moved Temporarily
Location: http://localhost:8000/baz/

Solution: Proxy Redirect ⌘

location /foo/ {
  proxy_pass http://localhost:8000/bar/;
  proxy_redirect on;
}

Adding an X-Forwarded-For ⌘

• The proxy destination doesn't know who made the original request
• Solution: Add an X-Forwarded-For header

location /foo/ {
  proxy_pass http://localhost:8000/bar/;
  proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
}

Dealing with legacy applications + cookies ⌘

• Scenario: multiple applications need to be on a single domain

• http://app.example.com/legacy1
• http://app.example.com/legacy2

• Problem: These applications set cookies for the domain they are on - not the new single domain

nginx to the rescue, again ⌘

location /legacy1 {
 proxy_cookie_domain legacy1.example.com app.example.com;
 proxy_cookie_path $uri /legacy1$uri;
 proxy_pass http://legacy1.example.com/;
}
location /legacy2 {
 proxy_cookie_domain legacy2.example.com app.example.com;
 proxy_cookie_path $uri /legacy2$uri;
 proxy_pass http://legacy2.example.com/;
}

What did we do? ⌘

• nginx will rewrite the cookie domain to app.example.com
• nginx adds a cookie path matching the URI
• Each app sees only its own cookies

Caching ⌘

• nginx can be used to accelerate content from the upstream
• By caching dynamically generated content, performance can be improved massively

Simple caching example ⌘

http {
proxy_temp_path /var/spool/nginx;
proxy_cache_path /var/spool/nginx keys_zone=CACHE:100m levels=1:2 inactive=6h max_size=1g;
 location / {
   proxy_pass http://localhost:8000/;
   proxy_cache CACHE;
 }
}

What we did ⌘

• Configured a 100MB in-memory cache
• Configured a 1GB on-disk cache
• Told nginx to cache everything suitable

What nginx does ⌘

• nginx proxies requests to localhost:8000 as in previous examples
• nginx looks at the headers received to check if data is safe to cache, and for how long
• nginx stores a copy on disk
• Next time a request is made for this URL, on-disk cache is checked
• If it exists and is fresh, serve directly, storing a copy in RAM

What is fresh? ⌘

• Backend server should send appropriate headers to let nginx (and any other caches along the way) what is OK to cache, and how long for
• `Cache-Control: private`

• don't cache at all

• Cache-Control: public, max-age: 900

• cache for 15 minutes

• Expires: Mon, 1 Jul 2013 01:00:00 GMT

• cache until the specified date

Cache stampedes ⌘

• Cache stampedes are a big problems for sites using caching to scale
• Imaging the following scenario:

• Proxy destination can handle max 50 requests/second
• nginx is added, traffic is 300 requests/second
• 25% of traffic is for the front page
• what happens when max-age is reached?

Dogpile ⌘

Dogpile effect ⌘

• Everyone tries to get the same URL at once
• It's very slow / broken for everyone until a new copy gets into the cache
• Load graphs show waves of peaks

nginx solution ⌘

• nginx can be configured to serve stale content whilst updating

location / {
  proxy_pass http://localhost:8000/;
  proxy_cache CACHE;
  proxy_cache_use_stale updating;
}

Hiding errors ⌘

• If old content is better than no content, nginx can also be configured to use stale content when it recieves no response or an error from the backend
• We also set a timeout value

location / {
  proxy_pass http://localhost:8000/;
  proxy_cache CACHE;
  proxy_cache_use_stale timeout http_500;
  proxy_read_timeout 15s;
}