[Linux-cluster] NFS failover on GFS?

ptr at poczta.fm ptr at poczta.fm
Fri Jul 15 11:04:19 UTC 2005


   Hello.

   As I need to incorporate failover on 2-nodes GFS cluster, I set the heartbeat communication between nodes exporting
GFS partition for external hosts. Although production tests 
brought somewhat unsuccessful results:

1. If I set an NFS as service run with heartbeat script (haresources), so it's being started and shut dwon togetheer with shared cluster interface, I'm getting problems with NFS sstartup on node taking resources. It works when juzt stopping heartbeat on one of the nodes, so another one takes resources over and starts services (NFS in this case). 
BUT: when I test ral-life scenario as by pulling the ethernet plug from one node, the shared interface is brought up on second one, but NFS won't start due to unknown reason.
The shared directory with NFS locks is /var/lib/nfs located on GFS partition, so I assume the moment of fencing dead node makes trouble for proper NFS startup on second node.

2. I also attempted to remove NFS script form services managed by heartbeat, but the same situation happens: although HA sends the gratituous arp when taking the shared interface over, machines mounting GFS partition won't see the change and will loose mounts.
   Any help on this issue would be greatly appreciated.
Regards,

Piotr

----------------------------------------------------------------------
Najnowsze wiadomosci!!! >>> http://link.interia.pl/f18a0





More information about the Linux-cluster mailing list