I just discovered pound a couple days ago, in a search to find a
solution for load balancing webservers. In my situation, I have an app
which picks the worst times possible to die, and it isn't something I
can modify the source on. For sure, if visitors get an HTTP 500 series
result, the app most certainly hung, and requests should be routed to
the next available server.
 
I have gone through the archives, and see this that there is a feeling
that HTTP 500 should not be used to determine server health. I'm here to
tell you that it most certainly should, at least in my case.
 
In my case, I have no control over the source code, and providing
service is the first priority. For me HTTP 500 = Service Unavailable. My
world would be so much better if pound would let me mark as server
returning HTTP 500 errors as dead, so it would skip it and use the next
available server.
 
It would be great if this was at least an option that could be set in
the .cfg file; "Offline on 500 err" or something like that.
 
Alan McCollough
 

you may want to consider HAproxy.  it has a much more flexible
health checking option.  look at the documentation for the "httpchk"
configuration option.

we looked at both pound and HAproxy very carefully when building out
our application stack and in the end, the health checking was the
one big thing that swung us towards HAproxy.

good luck
michael

--
Michael Lasmanis
mpl(at)corclix.com
__________________________________

Corclix, Inc.
2625 Middlefield Rd. #567
Palo Alto, CA 94306
__________________________________



On Sep 30, 2008, at 12:06 PM, McCollough, Alan wrote:

> I just discovered pound a couple days ago, in a search to find a
> solution for load balancing webservers. In my situation, I have an app
> which picks the worst times possible to die, and it isn't something I
> can modify the source on. For sure, if visitors get an HTTP 500 series
> result, the app most certainly hung, and requests should be routed to
> the next available server.
>
> I have gone through the archives, and see this that there is a feeling
> that HTTP 500 should not be used to determine server health. I'm  
> here to
> tell you that it most certainly should, at least in my case.
>
> In my case, I have no control over the source code, and providing
> service is the first priority. For me HTTP 500 = Service  
> Unavailable. My
> world would be so much better if pound would let me mark as server
> returning HTTP 500 errors as dead, so it would skip it and use the  
> next
> available server.
>
> It would be great if this was at least an option that could be set in
> the .cfg file; "Offline on 500 err" or something like that.
>
> Alan McCollough
>
>
>
> --
> To unsubscribe send an email with subject unsubscribe to pound(at)apsis.ch 
> .
> Please contact roseg(at)apsis.ch for questions.