[Linux-cluster] CLVMD without GFS

[Moralejo, Alfredo]

Hi,

By default, when clvmd service is restarted, all logical volumes in clustered filesystems are activated. I understand this is a problem to implement the exclusive activation, as it should work in the opposite way, all logical volumes should be deactivated until the resource taking care of it is started in a node where it'd be exclusively activated.

Has anyone found this problem while implemented the lvm-cluster.sh resource or a similar solution? Is there a way to fix it via lvm configuration or something like that?

Best regards,

Alfredo


-----Original Message-----
From: linux-cluster-bounces at redhat.com [mailto:linux-cluster-bounces at redhat.com] On Behalf Of rmicmirregs
Sent: Saturday, August 22, 2009 10:14 AM
To: linux clustering
Subject: Re: [Linux-cluster] CLVMD without GFS

Hi,

El mar, 18-08-2009 a las 16:19 +0200, brem belguebli escribió:
> Hi,
>  
> I see what you mean.
>  
> On the other hand, it would may be lessen some flexibility. The
> typical case being the fact that you do not want to use FS on top of
> your LV's but raw devices.
>  
> Brem
> 
>  
> 2009/8/18, Moralejo, Alfredo <alfredo.moralejo at roche.com>: 
>         Hi,
>         
>          
>         
>         Yes, I understand that. 
>         
>          
>         
>         However, what I meant is that, from a configuration simplicity
>         point of view, it could be convenient to add the functionality
>         into a only resource that take care of the entire cycle,
>         logical volume activation in exclusive mode and mounting it. I
>         think this is the most common usage of the fs resource. In
>         this way, I don't need to create two resources for each
>         filesystem (one for the logical volume and one other for the
>         fs) and I can be sure a filesystem can not be mounted by hand
>         in another node (as it has been activate exclusively in the
>         node running the service).
>         
>          
>         
>         Best regards,
>         
>          
>         
>         Alfredo
>         
>          
>         
>          

What Alfredo asks is exactly the point i wanted to reach at the end of
my lvm-cluster resource script: a filesystem resource with grants of
exclusive use, avoiding administrators make the mistake of, by hand,
mount concurrently the same FS on two different nodes.

As far as i know, and correct me please if i'm wrong, there is no
"exclusive usage" flag on non-clustered FS, even maybe nor in clustered
FS. So i moved the protection to a lower layer into the storage stack:
the LVM layer.

This approach is the same i was accustomed to use with the Linux-HA +
EVMS (Enterprise Volume Manager System) approach. I used to use
Clustered Segment Managers to build the /dev EVMS devices that were used
as mountpoints, so if an administrator wanted to mount the FS by hand he
was needed to assure the CSM was owned by the node he was working on. 

I agree with both of you that this would be much simpler is there was no
need of implicating LVM layer on the protection of the FS, but i have
not found any other way of doing this.

In the other hand, there is also the "bug" of the overlapping of the LVM
exclusive flag that Brem has opened in the bugzilla. If this gets fixed
in the way we think it should be, i think we would have reached a good
complexity/success ratio for this security solution.

Cheers,

Rafael


-- 
Rafael Micó Miranda

--
Linux-cluster mailing list
Linux-cluster at redhat.com
https://www.redhat.com/mailman/listinfo/linux-cluster