100% with upstream / backup server

[Maxim Dounin]

Hello!

On Tue, Jan 26, 2010 at 05:39:02PM +0100, Matthias Rieber wrote:

> Hi,
> 
> I've noticed a strange issue. I've a configuration like:
> 
> ...
> upstream _backend {
>   server 10.0.0.1:80 down;
>   server 10.0.0.2:80 down;
>   server 127.0.0.1:80 backup;
> }
> ...
> 
> 
> the purpose is to disable all backends and switch to some kind of
> maintenance mode.
> 
> Howerver, with this config, all workers have after the first
> requests a very high cpu load and nginx is not properly answering
> requests anymore.
> Whenn I kill the workers, they reappear with the high load.
> 
> I such a setup not allowed?

Yes, with all servers marked as 'down' nginx is caught in infinite 
loop in ngx_http_upstream_get_peer().  I'll take a look how to fix 
this properly.

Maxim Dounin