Re: Example "local" fails on node with two IP addresses

Hwllo!

On Mon, Apr 23, 2007 at 08:28:59AM +0200, Verdi March wrote:
> Apr 23 12:57:07 sun-n1-console kernel: Lustre: Added LNI 129.158.130.75@tcp 
> [8/256]

You should have included full log like this from the very beginning,

> 
> Apr 23 12:57:12 sun-n1-console kernel: LustreError: Refusing connection from 
> 192.168.123.45 for 192.168.123.45@tcp:  No matching NI
> Apr 23 12:57:12 sun-n1-console kernel: LustreError: 
> 4416:0:(socklnd_cb.c:2160:ksocknal_recv_hello()) Error -104 reading HELLO 
> from 192.168.123.45
> Apr 23 12:57:12 sun-n1-console kernel: LustreError: Connection to 
> 192.168.123.45@tcp at host 192.168.123.45 on port 988 was reset: is it 
> running a compatible version of Lustre and is 192.168.123.45@tcp one of its 
> NIDs?

> It looks to me that there's a confusion over which network interface
> to use (eth0 = 129.158.130.75, and eth1 = 192.168.123.45).

Right.

> I intended to deploy MDS on eth1; this is specified using IP address
> when creating a node:
>   --add net --node sun-n1-console --nettype lnet --nid 192.168.123.45@tcp

This won't help.

> Are there anything else that I missed?

Yes, you need to pass lnet module option 'networks' like this in your
/etc/modprobe.conf:
options lnet networks=tcp(eth1)

(naturally replacing eth1 with interface that has the address you want to listen
on)

Bye,
    Oleg