Hi. My partner (Ravi) and I are setting up two separate linux clusters but
we are having similar problems.
For this posting, I will only talk about one, but we need help solving the
problems for both :). Hopefully, it is the same problem and easily
diagnosed by someone with experience.
DR is being used.
The kernel is 2.4.5, with the ipvs-0.9.1, and hidden-2.4.5-1 patches
applied to it. Configure-0.8 was used to set this cluster up. Also, the
value in
/proc/sys/net/ipv4/conf/all/hidden is set to 1 for the real-servers and 0
for the director.
There is one director and 2 real-servers; set up in the following manner:
director:
VIP: arlx016
DIP: 10.1.1.1
real-servers:
VIP: arlx043
RIP: 10.1.1.2
VIP: arlx044
RIP: 10.1.1.3
The symptoms of the problem are that when requests are sent the director
never processes the packets (or at least tcpdump doesn't report them) and
hence the scheduling algorithm is never called. The number of connections
is not updated.
After pinging the director VIP, the arp -a on the client reports:
arlx016.austin.ibm.com (9.3.61.16) at 00:60:CF:20:5D:8F [ether] on eth0
which is the MAC address of a realserver (10.1.1.2)
pinging arlx016 from itself reports the following:
[root@arlx016 /root]# arp -a
? (10.1.1.2) at 00:60:CF:20:5D:8F [ether] on eth1
? (10.1.1.3) at 00:60:CF:20:5D:9C [ether] on eth1
So it seems that the real-servers are still responding to arp requests.
Another interesting problem is that although the lvs_dr.conf file only
mentions the service www and that is all that ipvsadm reports, all services
are distributed to the real-servers.
Any suggestions would be appreciated, please let me know if you need any
more information.
Thanks,
alison
|