I searched and only found it in the source, and there's no comment  
there as a hint. I don't grok C very well, but best I can tell this  
means all back-ends are either !alive or disabled. Apologies if the  
meaning of that is totally obvious.

My Alive value is 30 seconds.

Service block from my conf:

	Service
             DynScale 1
             BackEnd
                 Address 192.168.0.20
                 Port    80
		Timeout 300
             End
             BackEnd
                 Address 192.168.0.23
                 Port    80
		Timeout 300
             End
             Session
		Type    COOKIE
		ID      "PHPSESSID"
		TTL     86400
	  End		
         End

Thanks for your consideration.

- Tom

Hi Tom,

http://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html

HTTP code 503 is "Service Unavailable". This means that pound can not
connect to 192.168.0.20 on port 80.

This can mean many things:

1. No route to host
   /sbin/route -n (shows you the route table using IP and not names - much
faster)

2. Firewall on 192.168.0.20 is blocking inbound on port 80

3. Firewall on your pound is blocking outbound on 192.168.0.20:80

4. NIC hosting 192.168.0.20 is not alive
   /sbin/ifconfig (shows you the nic config for your box)

5. Cable is not attached

6. You are using a cross-over cable instead of a patch cable, check your
switch for the pretty green happy indicator.

7. Service at 192.168.0.20 on port 80 is not responding

7(a). Check that service on 192.168.0.20 can route back to your pound box.
Do an icmp ping on 192.168.0.20 from the pound box to make sure that you can
get to and from traffic to your BE.
[...]

Many thanks for the detailed reply, Jake. It looks like number 7 is  
the problem in my case. Even though the back-end servers aren't under  
heavy load (according to uptime), and there are no errors in their  
logs, they must have reached their limit. Apache tuning might help,  
but that's off-topic here.

The connection certainly is timing out before 300 seconds, but maybe  
this error means it's unable to even establish TCP/IP connection so it  
never starts waiting for the HTTP response. .... or maybe the timeout  
has already occurred and the back-end is disabled until the next Alive  
check.

I'm going to try throwing another back-end machine into the mix and  
see if this solves the problem. I'll let you know how that goes,  
though it may be a while before we see traffic like we've had recently.

Thanks again.

- Tom

On May 15, 2008, at 2:01 PM, Jacob Anderson wrote:
[...][...][...]