It is fixed. I replaced portmap on host with rpcbind and used "-i insecure"
mode which allows SET and UNSET from any host. Guests are served
by hosts rpcbind daemon.
rpcbind has -h option for binding to specific ip, but it binds to
0.0.0.0 anyway,
so the last possible combination finaly worked.
Thanks, the exchange helped me to get unstuck.
On Mon, May 17, 2010 at 3:31 PM, JF Straeten <jfstraeten@scarlet.be> wrote:
> On Mon, May 17, 2010 at 02:48:21PM +0200, Petar Hitij wrote:
>
>> > I think it's the opposite : NFS can't start because portmap isn't
>> > running...
>
>> portmap is running. It is not listening on 127.0.0.1 so other daemons
>> cannot register.
>>
>> I cannot get portmap to listen just to 127.0.0.1 and to eth0 interface
>> ip. It is one or the other.
>
> It was tricky, IIRC, but it works here on the host with just
>
> OPTIONS="-i 127.0.0.1"
>
> in /etc/default/portmap
>
> (And nothing in guests.)
Which version/distro?
>
>
>> > NB 1) out of memory : it you set "-i 127.0.0.1" for portmap on the
>> > host, IIRC you should specify the 'port' and 'mountport' options on
>> > the clients for the NFS mount to succeed (they can't contact the
>> > host portmap since it's not running on the network IP, just on the
>> > lo, so a hint is necessary)
>>
>> True, but not maintanable.
>
> Yes :-/
>
>
>> There is one more option - use one portmapper for all vservers.
>> It doesn't work because portmap is filtering register requests
>> from "non-local" ip numbers.
>
>
> Not tried... But BTW, there are also other options :
>
> - user space NFS server ? unfs3 can work without portmap to some
> extend (but will require port/mountport on clients too, so perhaps not
> an option in your case...) ;
Have tried it and came to the same conclusion.
>
> - glusterfs ? It's a user space cluster FS, easy to setup and with
> very decent performances.
I have Sun (oracle?) machines for clients...
>
>
> Hih,
>
>
> --
>
> JFS.
>
Received on Mon May 17 14:57:08 2010