|
I'm using lvs to load balance several hundred ssh and telnet sessions.
The problem I'm having is my ipvsadm table does not show the actual
number of connections.
For example:
TCP cvg1-app:ssh wlc
-> fc-app-cvg1-104.cvg1:ssh Route 10 27 0
-> fc-app-cvg1-102.cvg1:ssh Route 10 25 1
Using just these two servers as an example (there are several other
real servers), it shows 27 and 25 active connections. However:
fc-app-cvg1-102.cvg1$ w
17:10:47 up 48 days, 13:22, 98 users, load average: 3.33, 3.37, 3.28
fc-app-cvg1-104.cvg1$ w
17:11:03 up 21 days, 17:13, 96 users, load average: 0.40, 0.54, 1.26
fc-app-cvg1-104 has 96 users, but lvs only shows 27 active connections.
fc-app-cvg1-102 has 98 users, but lvs only shows 25 active (1 inactive).
I've increased all of the timeout_* values in /proc/sys/net/ipv4/vs/
to 28800
That did solve the problem of some telnet sessions closing due to a
relatively short period of inactivity, however it does not address the
problem of the value of active connections being grossly different
than the number of logged in users.
I have verified that all authenticated users have logged in via the
vip. Some of these users have been logged in for several hours, some
over 24 hours. Their connections are still active, even though they
may have had several hours in inactivity.
I don't understand at all why the numbers are so different, yet the
connections are still active. Any insight to this mystery would be
greatly appreciated.
Regards,
Bill Omer
|
