|
After a lot more research, I'm leaning towards blaming my problems on the 15
minute session timeout LVS seems to impose. This has been a problem for
people in the past, but unfortunately all the solutions seem to involve
changing the tcp timeout value via "ipchains -M -S x x x"..... and I'm using
netfilter, where ipchains doesn't work.
What's confusing me is that the default tcp timeout with netfilter is
supposed to be 2 hours, and yet "ipvsadm -lcn" shows sessions as having a 15
minute TTL. How do I increase that value? We're keeping tcp sessions open
for hours at a time, so 15 minutes isn't working well for us.
----- Original Message -----
From: Ben
To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Sent: Thursday, August 15, 2002 10:51 AM
Subject: connection wierdness
We're developing a server which we would like to place behind an LVS machine
for automatic failover, and we're noticing some wierd problems with our
clients disconnecting. In trying to figure out why, I'm seeing that
connections which should be up (but seem to have been dropped, according to
our clients and server) are listed in /proc/net/ip_conntrack, but not
/proc/net/ip_vs_conn. I was under the impression that all lvs-managed
connections lived in latter and not the former, but is that not the case?
|
