Hi all, I recently put 1.6.0.1 on debian etch using sources from: http://www.pdsi-scidac.org/repository/debian/index.html Hardly any testing but all appears quite kosher. A question regarding failover. Previously, on 1.4.x, we specified failover pairs and groups and generated configuration files which lconf used. Now, we simply do a mount -t lustre (...) and we''re good to go. Am I correct in assuming that any OSS can simply mount the storage device in question? Regards, mustafa. -- Mustafa A. Hashmi mahashmi@gmail.com
Mustafa A. Hashmi wrote:> Hi all, > > I recently put 1.6.0.1 on debian etch using sources from: > > http://www.pdsi-scidac.org/repository/debian/index.html > > Hardly any testing but all appears quite kosher. > > A question regarding failover. Previously, on 1.4.x, we specified > failover pairs and groups and generated configuration files which > lconf used. Now, we simply do a mount -t lustre (...) and we''re good > to go. > > Am I correct in assuming that any OSS can simply mount the storage > device in question?Mostly, yes. From the HA perspective, Lustre can now be managed just like any other filesystem, if you are using Heartbeat you can now use the built-in Filesystem resource. On the Lustre setup, there are two things that may need to be done: MDS failover: When creating the filesystem, a pair of MDS nodes should be specified with the --mgsnode= parameter. ( And clients must reference both MDS NIDS when mounting, this is unchanged from 1.4 ) OSS failover: When creating, specify the failover partner with the --failover= parameter. These parameters can be changed after filesystem creation with tunefs.lustre. cliffw> > Regards, > mustafa.
On 7/13/07, Cliff White <cliffw@clusterfs.com> wrote:> Mustafa A. Hashmi wrote: > > Hi all, > > > > I recently put 1.6.0.1 on debian etch using sources from: > > > > http://www.pdsi-scidac.org/repository/debian/index.html > > > > Hardly any testing but all appears quite kosher. > > > > A question regarding failover. Previously, on 1.4.x, we specified > > failover pairs and groups and generated configuration files which > > lconf used. Now, we simply do a mount -t lustre (...) and we''re good > > to go. > > > > Am I correct in assuming that any OSS can simply mount the storage > > device in question? > > Mostly, yes. From the HA perspective, Lustre can now be managed just > like any other filesystem, if you are using Heartbeat you can now use > the built-in Filesystem resource. > > On the Lustre setup, there are two things that may need to be done: > MDS failover: > When creating the filesystem, a pair of MDS nodes should be specified > with the --mgsnode= parameter. ( And clients must reference both MDS > NIDS when mounting, this is unchanged from 1.4 ) > > OSS failover: > When creating, specify the failover partner with the --failover= parameter. > These parameters can be changed after filesystem creation with > tunefs.lustre.Thank you Cliff. I didn''t catch this in the manual -- will look again. -- Mustafa A. Hashmi mahashmi@gmail.com
Hi Cliff, On Friday 13 July 2007 00:30:07 Cliff White wrote:> On the Lustre setup, there are two things that may need to be done: > MDS failover: > When creating the filesystem, a pair of MDS nodes should be specified > with the --mgsnode= parameter. ( And clients must reference both MDS > NIDS when mounting, this is unchanged from 1.4 )this part I didn''t understand entirely. Its clear one has to specify several mgsnodes on creating an OST, but on creating the mgsnode, does one have to say "--mgs --mgsnode={failover_system}"? Thanks, Bernd -- Bernd Schubert Q-Leap Networks GmbH