On Thu, 2009-04-02 at 13:06 -0500, Anoop Bhat wrote: > The servers are Centos 5.2 running piranha.
OK... so this question may be better off being asked of the Piranha mailing list, which has been mentioned on numerous occasions on this list. Relatively few subscribers to this list use Piranha so far as we can tell. > We're wondering if there are connections that make it to LVS that somehow > don't get redirected to the real servers. Possibly, but it depends on what Piranha (and the constituent parts thereof) are doing. > Is there a way to determine if this is happening and also to gather any > metric information on these occurrences. There might be, but with 10000 clients you *really* don't want to have to switch on LVS debugging, believe me! It's also very likely that the CentOS kernel is not compiled to support LVS debugging. You'll probably end up looking at the bits of Piranha that do health checking. As a non-user, I have no idea what they are - mon, perhaps, along with heartbeat for failover? I'd guess you can switch something on there to produce debug logs. At a more basic level, run tcpdump on a few of the clients (or the equivalent) and also on the directors and realservers, filtering for the client source addresses. If you're able to keep all the clocks in sync and have enough space to store the pcap files, you should be able to trap enough data to correlate where the traffic is disappearing. Also worth considering is whether or not you have the netfilter (iptables) conntrack modules loaded somewhere with a table size set too small... that might cause connections to be flushed from an ESTABLISHED state before they really are finished, thus causing the connection to be dropped. Graeme _______________________________________________ Please read the documentation before posting - it's available at: http://www.linuxvirtualserver.org/ LinuxVirtualServer.org mailing list - [email protected] Send requests to [email protected] or go to http://lists.graemef.net/mailman/listinfo/lvs-users
