|
I want to use LVS with DR. All worked fine on first tests. But if I
stress LVS with 500 TCP-connectiions per second, I can see a big memory
leak. The server is leaking about 1GB in a day! It is not possible to
freeing the laked memory without a reboot. The memory leak is in the
slab. "size-2048" is growing.
I am using a SLES Kernel, and I tried 2.6.5-7 and 2.6.16-21. Same leak
in both cases!
I have made further trails. It's very easy to reproduce the failure with
the newest 2.6.19.1 kernel. To see the leak I do a
while true; do
wget -nv -q -t 1 -T 3 -O - http://VIP/test.gif >/dev/null 2>&1
done
on the client. This loop is generating about 60 connections per second.
Parallel I watch the size of the "size-2048" container in the slabtop
tool of the director machine. It is growing with about 2K per second!
This is my test configuration:
# ipvsadm -S
-A -t VIP:http -s wlc
-a -t VIP:http -r REALSERVER1:http -g -w 1
# ipvsadm -L --timeout
Timeout (tcp tcpfin udp): 900 5 20
I can not believe that there is a so obviously bug in LVS. Is there
something wrong in my configuration? How can I isolate the reason of the
memory leak?
Winfried
|
