View Full Version : Can't export NFS in Fedora 9 x64 Virtual Cluster
dzspb
2nd May 2009, 11:39 AM
Hi, I'm trying to get experience in creation of virtual Fedora 9 x64 cluster using Ubuntu 9.04/Virtual Box 2.2.2. I have a head node with 1 G of RAM and two slave nodes with 512 M each. There's internal networking in Virtual Box with DHCP, so my nodes have addresses: 192.168.15.100/24 (Slave0.localdomain), 192.168.15.101/24 (Slave1.localdomain), 192.168.15.102/24 (Head.localdomain). Ping is OK. I've added the following string to /etc/exports file: /usr/share 192.168.15.0/24(rw) And then used /usr/sbin/exportfs -a But showmount -e 192.168.15.102 entered at slave node shows nothing. What's wrong? May be I have to somehow start nfsserver, but how?
stevea
2nd May 2009, 12:06 PM
I don't have an F9 handy and some of the services names have changes between F8 & F10.
On the server you absolutely need to have rpcbind, rpcsvcgssd and rpcidmapd running. Some of these are started by the "nfs" service (which is only needed by the server).
On the client you should have rpcgssd and rpcidmapd running.
You can manipulate the services from the menu (System->Admin...->Services) or using the same tool from the command line as "system-config-services" or you can manipilate individual services (as root) like "service nfs restart" "service rpcgssd start" and so on.
After you do the "exportfs -a" on the server, then do a "exportfs -av" to see if the share created OK.
Also on the server try this (as root)
mkdir /tmp/mp
mount -t nfs Head.localdomain:/usr/share /tmp/mp
ls - l/tmp/mp
...
umount /tmp/mp
You should be able to mount the share locally and access it (but make sure rpcgssd is running on the server if you want access permission to work).
FWIW I think NFSv4 is substantially superior to NFSv3 and the system-config-nfs can help but it's poorly documented. To do this make an /etc/exports file like
/usr/share 192.168.15.0/24(rw,fsid=0)
Then the mount command looks like:
mount -t nfs4 Head.localdomain:/ /tmp/mp
*** Note that on NFSv4 you only spec the root not the server path
This "server:/", not this "server:/usr/share"
You can still mount the share w/ the V3 command but it then uses the slacker and slower V3 protocol.
dzspb
2nd May 2009, 02:13 PM
Thank You, Stevea, now I can export nfs and mount it locally. But I still can not mount it from another node. May be I need enable some additional services? The error message is 'mount.nfs: internal error' on enter: mount -a -t nfs when the last line in /etc/fstab is: 192.168.15.102:/usr/share /head/usr/share nfs rw,bg,intr,hard 0 0 The directory tree /head/usr/share I've created allready.
stevea
2nd May 2009, 05:19 PM
Sorry - thought I spotted an error , but not so.
Have you tried mounting it manually and messing around with the options ?
I just tried your fstab options and command - and it mounts fine on F8.
Verify you can manually mount.
Firewall ?
Perhaps pull up wireshark and review the traffic ?
dzspb
2nd May 2009, 08:34 PM
When I enter mount -t nfs 192.168.15.102:/usr/share /head/usr/share in terminal on slave node I get 'mount.nfs: internal error'. May be it's so because there's no static IP lease in my VirtualBox DHCP-server used. May be - firewall, but I don't know where can I disable it. This - as for my virtual cluster. And I have two PS3-boxes running YDL 6.1, a kind of Fedora, but with a poor UI. If I'd know how to export and mount nfs and enable suitable services using command-prompt in terminal only, I would try to do this exercise on physical cluster of two PS3 tomorrow.
dzspb
3rd May 2009, 07:50 AM
I couldn't mount from slave node because at head node in file exports there were additional key - insecure_locks, i.e.
(rw,insecure_locks). Without it all works fine.
dzspb
19th May 2009, 01:11 PM
Stevea, I tried to reproduce my success with NFS on two PS3's running YDL (Fedora-based linux). I've got two error messages, the first on a server and on a client:
# service rpcbind restart
rpcbind: unrecognized service
And the second - on a client:
#showmount -e 192.168.12.243
mount clntudp_create: RPC: Port mapper failure - RPC: Unable to receive
The result of exportfs -av on server is OK. What's wrong?
stevea
20th May 2009, 05:16 AM
Odd - it's called "rpcbind" in F8 and F10. Either it's not installed or not installed correctly. Try a "yum erase" and yum install" for the "rpcbind" package.
Do you have a file /etc/init.d/rpcbind ??? If so perhaps try "chkconfig rpcbind on". Then see the the 'service' command works.
The showmount failed "portmapper failed" b/c the portmapper service (rpcbind) isn't running.
vBulletin® v3.8.7, Copyright ©2000-2013, vBulletin Solutions, Inc.