Most of the LAMP clusters I deal with on a daily basis use the same basic stack – an nginx at the front as a load balancer proxy, apache as the application server (running php), and mysql as the backend. I just realized that this stack might as well be called LNAMP or LAMPN, since nginx plays a big part in it.

Problem with LAMPN, is that it sometimes takes its new stack name seriously, and starts limpin’. Nginx is an excellent front end – it’s highly configurable and highly efficient. But, it has a nasty habit of not letting your requests go through to the application servers if the request is hard for it to understand. Instead, it returns a “400 Bad Request” and dies.

So what was a request that was “hard to understand” in my case? It was a request with a very large “Cookie” header. If the cookies size of a certain domain sum up to more than large_client_header_buffers, your request is simply rejected by nginx. The default for large_client_header_buffers is:

large_client_header_buffers 4 4k 
/* 4k being the page size of the system, can be any size depending on OS */

The cookie header sent from my browser to the domain I tried to access was 4.4k in size, larger than the default size, so nginx flipped. I read it is only a Firefox issue (does IE split the Cookie to chunks? Is it even possible to send multiple chunks of a Cookie header?), but it might as well happen with other browsers and different requests. To solve the problem, the following setting in the http context will solve your problem:

large_client_header_buffers 4 8k

Actually, nginx documentation mentions this problem, but it’s one of those default settings you think to yourself you’ll never have to change. Well, apparently, sometimes you do.