Re: clustering resin servers

On Jun 28, 2006, at 12:22 PM, Eric Kreiser wrote:

> I have a cluster defined with 2 srun's... 1 as a primary and 1 as a  
> backup... my issue is that when my primary goes down... the next  
> request fails while the second request correctly goes to the backup  
> srun.  Why isn't the first request correctly sent to the backup?

Well, this is a reason we'll start encouraging people to use Resin as  
the web tier.  (Plus we're improving the load balancing in 3.0.20).

The issue is that the web-tier doesn't know that the backend has  
failed until it gets a failure from the write() call.

A reasonably-sophisticated load balancer (like the Resin one) can  
retry the same request on a second backend.   mod_caucho can't.  (And  
we will not be adding that capability to mod_caucho because it's not  
as testable as Resin is.)

Actually, there are a number of advantages of using Resin in the web  
tier.

   * Resin can already act as a proxy cache, so static or pseudo- 
static files can be served directly from cache.

   * Because Resin is fully threaded, it can share information and  
sockets and caches more easily than Java can.

   * Resin's load balancer can do some clever things like warmups (in  
3.0.20) and more effectively dealing with updates (probably 3.0.21).

-- Scott

>
> I front end resin with apache 2.0.55.
> Running resin 3.0.14
>
>
>