[Linux-cluster] sharing attr on clustered volume -- cannot see open attribute

[Fabio M. Di Nitto]

On 10/14/2011 02:38 PM, Daniele Palumbo wrote:
> Il giorno 14/ott/2011, alle ore 14.01, Fabio M. Di Nitto ha scritto:
>> This setup looks very wrong and there is a lot of work on the storage
>> side you need to do.
>>
>> I am not even sure where to start, but a few simple points:
>>
>> 1) you need to use proper shared storage. AOE is fine, but use the real
>> one. not 2 local disks exported, because that's never going to work.
> 
> why not?

this is the kind of answer that boils down to RTFM.....

more seriously, the technical explanation is very long and complex.

> 
>> 2) if you want to use clvmd, all nodes *must* see the same storage
> 
> that is :)
> different name (in pvs) but same storage.
> anyway, i will add a third machine and setup the storage over there, then i will setup 3 machines in cluster that see the same device name.
> 
> can that help?

Yes, keep the 3rd machine out of the cluster and use it to export an AOE
device for testing.

> 
>> 3) you need a cluster filesystem such as GFS2. ext3 and reiserfs are not
>> cluster fs. If you mounted any of those on both nodes, I strongly
>> recommend recreateing the fs and restore data from a backup.
> 
> yes but that could be a problem... cause i need to have (at least) a volume for each virtual machine.
> or am i missing something like do a cluster filesystem and on top of that re-create the vg?

You were talking about mounting reiserfs and such... as long as you make
sure that the fs is mounted on one node at a time then it's fine.

No you don't need to create a vg on top of gfs2. gfs2 is not so
different from any other filesystems, except that you can mount it and
use it simultaneously on all nodes read/write.

> 
> but anyway, i just need to see the filesystem as open...
> do i need a clustered one to see the correct open attribute with lvs?
> how does the open attribute is set?

Ok can you please explain better what you mean here??

clustered lvs are available on all nodes at the same time. You probably
see a metadata sync issue due to "interesting" storage.

Fabio