Lustre discuss - May 2006 - suggestions for MDS and OST failover

On Apr 20, 2006  20:13 -0700, Bala wrote:
>   from documentation I understand that, to have 
> failover configuration, need to maintain active 
> server configuration using HTTP server and wget, 
> LDAP, or shared storage.
> 
> since we don''t have any other machines to serve
> as a HTTP, LDAP server and NFS server. (we have 
> our own failsafe product to take care of STONITH 
> issue and failover of machines)
Since 1.4.2 and later, you do not need an external system for maintaining
the active server.  Clients will try all "--failover" servers defined
in
the configuration for a given target.  Since 1.4.6 it is also possible to
specify multiple MDS servers for zconf mounting, for example:

mount -t lustre mdsnode1,mdsnode2:/mdsname/client /mnt/lustre

In the fstab "mdsnode1,mdsnode2:/mdsname/client" would be the
"device".
You definitely DO NEED STONITH to kill the primary server if doing
failover to a backup.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

thanks Andreas, for the information, then I will
try to install with out separate server to keep
the active server configuration.

-bala-


--- Andreas Dilger <adilger@clusterfs.com> wrote:
> On Apr 20, 2006  20:13 -0700, Bala wrote:
> >   from documentation I understand that, to have 
> > failover configuration, need to maintain active 
> > server configuration using HTTP server and wget, 
> > LDAP, or shared storage.
> > 
> > since we don''t have any other machines to serve
> > as a HTTP, LDAP server and NFS server. (we have 
> > our own failsafe product to take care of STONITH 
> > issue and failover of machines)
> 
> Since 1.4.2 and later, you do not need an external
> system for maintaining
> the active server.  Clients will try all
> "--failover" servers defined in
> the configuration for a given target.  Since 1.4.6
> it is also possible to
> specify multiple MDS servers for zconf mounting, for
> example:
> 
> mount -t lustre mdsnode1,mdsnode2:/mdsname/client
> /mnt/lustre
> 
> In the fstab "mdsnode1,mdsnode2:/mdsname/client"
> would be the "device".
> You definitely DO NEED STONITH to kill the primary
> server if doing
> failover to a backup.
> 
> Cheers, Andreas
> --
> Andreas Dilger
> Principal Software Engineer
> Cluster File Systems, Inc.
> 
> 

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com

Hi All,
       we would like to test out lustre in 4 nodes

  2 nodes for MDS active-passive configuration
  2 nodes for OST active-active configuration and
  few clients.

all 4 machines will be using external RAID array 
connected to the machines.

  from documentation I understand that, to have 
failover configuration, need to maintain active 
server configuration using HTTP server and wget, 
LDAP, or shared storage.

since we don''t have any other machines to serve
as a HTTP, LDAP server and NFS server. (we have 
our own failsafe product to take care of STONITH 
issue and failover of machines)

need to your suggestions on the following

1. For this kind of setup where to store active
server configuration? 

2. can I use LDAP server say on both the MDS servers
using one virtual IP, so that if one machine fails
other will take over, may need to sync the LDAP
servers
often.

3. can you suggest me is it ok, also let me know if 
your have any other better way to do that?

thanks but excuse me for long mail,
-bala-

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com