LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: IPVS Synchronization bug

To: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: IPVS Synchronization bug
From: Wensong Zhang <wensong@xxxxxxxxxxxx>
Date: Mon, 22 Oct 2001 09:49:43 +0800 (CST)
Hello,

On Sun, 21 Oct 2001, Fabrice wrote:

> Hello,
>
> >We use UDP multicast. The source address of UDP packet isn't that
> >important. In the code, I didn't bind the sending socket with a local
> >address specifically, the address:port (such as 192.168.1.11.1028 in your
> >example) is automatically allocated by the kernel.
>
> I think I didn't explain my problem enough clearly. The synchronization
> works when I use eth0 but stops working with eth2. Tomorrow I'll check if it
> works with eth1. (And yes, I configured the interface on both daemons)
>
>
> >The ipvs sync code set the specified interface for outgoing multicast
> >messages in the master side, and the specified interface for incoming
> >multicast messages.
> >If you set the master director send mulitcast sync info through eth0, and
> >the backup director listen to eth2 for mulitcast, then the receive queue
> >of the backup will be definitely empty.
>
> No, I didn't configure the master on eth0 and the backup on eth2. I always
> configured both daemons on the same interface.
>
> When both daemons are using eth2, the "connections informations" packets
> are correctly generated and sent trough eth2. The problem is that the queue
> "sock->sk->receive_queue" on the backup daemon is always empty.
>

I tried to reproduce your problem, but failed.

On the backup director, I started two tcpdump program

[root@koala /root]# tcpdump -n -i eth0 ip host 224.0.0.81

[root@koala /root]# tcpdump -n -i eth1 ip host 224.0.0.81


1) The first test is

[root@koala ipvs]# ipvsadm --start-daemon backup --mcast-interface eth0

The only "tcpdump -n -i eth0 ip host 224.0.0.81" shows that it receive
multicast packets, and the "ipvsadm -lcn" shows that new connection
entries are created in the kernel of the backup.

[root@koala ipvs]# ipvsadm --stop-daemon


2) The second test

[root@koala ipvs]# ipvsadm --start-daemon backup --mcast-interface eth1

The only "tcpdump -n -i eth1 ip host 224.0.0.81" shows that it receive
multicast packets, and the "ipvsadm -lcn" shows that new connection
entries are created in the kernel too.

[root@koala ipvs]# ipvsadm --stop-daemon


(Note during the tests, I started a client program on the other machines
to always access the LVS service, so that there will be new connections
needing synchronization)

So, there is no problem for the receiving daemon to listen to different
interface. I guess that there is something wrong in your configuration.

Regards,

Wensong


> When both daemons are using eth0, the "connections informations" are
> generated and sent as correctly as with eth2 but the queue receives the
> datas.
>
>
> Thanks,
>
> Fabrice Bucher
>



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