|
Jeff Amaral wrote:
>
> After some period of inactivity, I'll try to connect via a web browser
> to the virtual IP address, and one of two things will happen:
> 1. I'll get an error (in Mozilla, anyway) saying 'The document contains
> no data.' and the request will fail. Subsequent connection attempts
> will work fine, but only appear to connect to node 2.
> 2. I won't get an error, but I'll only get the second node in the
> cluster, and all subsequent requests will only go to node 2.
it's likely the arp problem
> Here's my only clue:
>
> If I run the ipvsadm command when load balancing is working correctly
> and I'm running apache bench a computer from outside the cluster, I'll
> see something like this:
> ----------------------------------------------------------------------
> las1:~ # ipvsadm
> IP Virtual Server version 1.2.0 (size=4096)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn InActConn
> TCP las-cluster.mos.org:http wlc
> -> las1.mos.org:http Local 1 10 2700
> -> las2.mos.org:http Route 1 9 2616
> ----------------------------------------------------------------------
fine.
> When load balancing is not working and I run the same apache bench
> command again, I'll see this:
> ----------------------------------------------------------------------
> las1:~ # ipvsadm
> IP Virtual Server version 1.2.0 (size=4096)
> Prot LocalAddress:Port Scheduler Flags
> -> RemoteAddress:Port Forward Weight ActiveConn InActConn
> TCP las-cluster.mos.org:http wlc
> -> las1.mos.org:http Local 1 0 0
> -> las2.mos.org:http Route 1 0 0
> ----------------------------------------------------------------------
no packets are being sent to the director
> There'll be no active connections, even to node 2! But apache bench will
> be running and returning data just fine (but only from node 2).
the connection requests from the client are going directly to node2.
Thanks for the nice description.
The most likely cause is that the director is not replying to arp requests
and node2 is.
There has been some discussion in the last week or so (see the mail archives)
about ldirectord getting confused about setting the arps etc.
Does the problem only arise following a director failover?
However no matter what, the realservers should not reply to arp requests, so you
should never connect to the realserver directly no matter how confused the
directors
are.
Joe
--
Joseph Mack PhD, High Performance Computing & Scientific Visualization
LMIT, Supporting the EPA Research Triangle Park, NC 919-541-0007
Federal Contact - John B. Smith 919-541-1087 - smith.johnb@xxxxxxx
|
