Hi,
Hayden Myers wrote:
We've been using the lvs project for loadbalancing for quite some time
now. We've experienced minimal problems until yesterday. For some reason
our director starts routing traffic to only one of the real servers in our
pool. ipvsadm output show almost all 0's for all servers. Normal numbers
Zero's for what? Active connections? Could you please send an ipvsadm -L
-n before and after this problem occurs?
are around 400 active connections for each active real server. We were
using an older 2.2.16 lvs kernel, and updated the system to a 2.2.19
kernel with ipvsadm 1.0.8 yesterday. We did recently move our realservers
to a 2.4 kernel which may be something to look at. We use the direct
routing method and the 2.4 kernels are patched with julians hidden patch.
Is it enabled? (Sorry to ask this, but it sounds like you've an arp problem)
Nothing but rebooting the director seems to get rid of the problem.
That I doubt, but lets have a clean analysis.
Clearing the ipvsadm table and reloading doesn't do much. I just turned
up the debug_level to see if anything pops up. Has anybody seen this
before? Any suggestions on where else to look?
Can you send a tcpdump trace of a session, please? Maybe if you already
enabled debugging, you could send a part of the output (the one, when a
session is initiated)
Regards,
Roberto Nibali, ratz
|