Nginx + fair load balancer patch looping

[Igor Sysoev]

On Fri, Feb 29, 2008 at 10:39:24AM +0100, Alexander Staubo wrote:

> An Nginx instance suddenly started spewing the following to its error
> log at a rate of about 1GB/minute, and using a bit more CPU than
> usual:
> 
>   2008/02/29 10:33:47 [error] 16875#0: *126309461 upstream prematurely
> closed connection while reading response header from upstream [...]
> 
> Aside from the excessive logging, everything else seemed normal. Our
> next_upstream setting is:
> 
>   proxy_next_upstream error invalid_header;
> 
> Restarting Nginx fixed the problem.
> 
> Could this be the fair load balancer going hairwire?

nginx logs this when upstream has closed connection before sending anything
or before it send full HTTP response header.

This is may be nginx bug, upstream failure, or kernel bug.

Could you truss failed process in next time ?


-- 
Igor Sysoev
http://sysoev.ru/en/