|
Hi
Just to add to Joe's comments...
On Mon, 2008-03-10 at 10:39 +0000, Dave Whitehouse wrote:
> Just looking for some advice around problem solving my LVS cluster
> performance.
>
> I have a cluster that currently has two LVS Directors hosting 4 VIPS. I use
> keepalived 1.1.13 on the directors, running on CentOS 4.4. Real servers are
> all CentOS 4.4 too.
CentOS 4 is a bit behind the bleeding edge but that does give it the
benefit of being somewhat more stable than, for example, Fedora :)
> I have hit an issue where my directors appear to be dropping packets. My
> connection rate is low when this occurs, typically about 40+ InPPS which is
> very low :-(
That is fairly low.
> A packet trace appears to indicate that packets are hitting the director but
> not coming out again.
As in, not being forwarded to the realservers?
> I have iptables running on the directors and ip_contrack is loaded along
> with a few other iptable related kernel modules. My question is do you thing
> that iptables could be causing problems with my LVS and if so have you any
> suggestions where and how I could prove it ?
Well, the easiest thing to do is to remove all the iptables rules and
unload the modules, then test. If the same problem occurs, you've found
that it isn't iptables and conversely if it doesn't, then it is.
You can then build your ruleset back up bit by bit and see which part
causes the problem.
For us to help you further we probably need to see your keepalived
config, the output from ipvsadm in normal operation and the output from
ipvsadm when there's a problem.
Also, have you "cloaked" the realservers to prevent them repsonding to
ARP requests for the VIP?
Graeme
|
