Upstream fail over?

[Marcus Clyne]

Hi Merlin,

It appears you're trying to do :

HTTP > Nginx > Varnish > (back to Nginx as failsafe) > application.

Apart from the fact that try_files won't work with the config you 
mentioned, I think you should probably be trying to use a different 
application stack (or at least use them in a different way).  One of the 
below options are probably better:

HTTP > Varnish > application         (Varnish has its own load-balancing 
and failover for connections to backends)
HTTP > Varnish > Nginx (load-balancing) > application
HTTP > Varnish > HAProxy > application      (HAProxy will typically 
out-perform Nginx when all you need to do is to load-balance)
HTTP > HAProxy > Varnish [> HAProxy] > application   (Very-high load and 
failsafe - only need initial load-balancing with a very large number of 
requests per second)
HTTP > Nginx (cache) > application      (use the proxy cache feature in 
Nginx, proxying to the application with a cache-miss)

If you have a normal load, then probably the first or the last of these 
suggestions would probably be best for you.  If you will have more than 
a couple of application servers, then I would probably recommend the 
last (i.e. using Nginx to cache, since I believe Nginx performs 
load-balancing better than Varnish at this time, though you can do more 
with VCL than you can with Nginx's conf files. 

Hope this is useful,

Marcus.


>   
>> merlin corey wrote:
>>     
>>> Hello,
>>>
>>> Yes, very easily with 0.7.x and above.  It would be something like
>>> below ( check http://wiki.nginx.org/NginxHttpCoreModule#try_files )
>>>
>>> location / {
>>>   try_files @varnish @application;
>>> }
>>>
>>> location @varnish {
>>>   // proxy to varnish
>>> }
>>>
>>> location @application {
>>>   // proxy to application
>>> }
>>>
>>> -- Merlin
>>>