LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: After failover, why VIP is left in primary server?

To: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: After failover, why VIP is left in primary server?
From: "Sachie Shinohara" <sshinohara@xxxxxxxxx>
Date: Wed, 24 Jan 2001 15:23:50 +0900
Thank you for helping me.

> Please explain your methodology in performing your failover tests.

In failover tests, under the both primary and secondary Piranha servers are
running, I down pulse deamon in primary server (#/etc/rc.d/init.d/pulse
stop).
Few seconds later (our lvs.cf set timeout 10s), secondary piranha server
takes over.

                                              |------------------------\
                                              |                                     \
                                    eth1:1  f.f.f.f                               Client PC
                                /                            \
                              /                                \
                            /                                    \
            eth1  b.b.b.b                               eth1  d.d.d.d
          |----------------- |                     |---------------------|
          | primary piranha     |                     | secondary piranha       | 
          |----------------- |                     |---------------------|
            eth0  a.a.a.a                                eth0   c.c.c.c
                           \                                 /
                             \                             /
                               \                         /
                                   eth0:1   e.e.e.e
                                             |
                                             |
                                            / \
                                          /     \
                              Web1           Web2
                       (real server)       (real server)

In this test after secondary piranha server took over pulse deamon, primary
piranha server still had both eth0:1 and eth1:1 VIP addresses (here, e.e.e.e
and f.f.f.f). Of course secondary piranha server had both VIP addresses,
too. As I could see the web page(IP was f.f.f.f) from client PC, pulse
wasn't dead. Ping to eth0:1 and eth1:1 from both primary & secondary was
fine.

What I want to know is this is normal or not?
> >I'm using linux-2.2.18 kernel, ipvs-1.0.2-2.2.18, and piranha-0.4.17-6.

> Depending on how you performed the failover test, this is normal. If you
> cut all paths of communication between primary and secondary, then both
> servers will have the VIP addresses.

I'm sorry, but I don't get this. In failover system, could it be possible to
cut all paths of communication between primary and secondary? When I used
"pulse" in piranha, I think both primary and secondary machines should  be
communicated  to each other, cause "pulse" works just as automated process.

This is the point.
> > One thing I noticed was when stopping piranha, pulse was sending this
> > message. 'pulse[****] : running command "/sbin/ifconfig" "eth1:1" "up"'
> > and 'pulse[****] : running command "/sbin/ifconfig" "eth0:1" "up".'
> > I am sure this message ran on stopping piranha.
I think "pulse" works wrong. When stopping piranha, message sould be like
"/sbin/ifconfig" "eth1:1" "DOWN"!!

Am I right?

Sachie

<Prev in Thread] Current Thread [Next in Thread>