Nginx + fair load balancer patch looping

[Grzegorz Nosek]

Hi,

On czw, cze 05, 2008 at 08:33:36 +0200, Andy Verprauskus wrote:
> I ran into this problem also; same error about upstream mongrels and 
> rapidly increasing log file size.  nginx was using as much cpu as it 
> could.
> 
> In our case, the particular upstream mongrel that was trying to be 
> contacted had died.  A few minutes after restarting the mongrel the 
> problem stopped.

Could you please contact me off-list (root&localdomain,pl)? I'd like to
know more about it because a backend simply dying should not confuse the
load balancer (it actually did at one time though, so please make sure
you're running the latest snapshot).

Also, as a totally wild guess (both for you and Alexander), could you
try increasing the shared memory segment size? I.e. try putting

upstream_fair_shm_size 64k;

inside your http { } section. If the fair balancer runs out of shared
memory, it logs a message to that effect but it may have easily been
lost in the torrent of later messages.

Best regards,
 Grzegorz Nosek