[Linux-cluster] CLVM in a 3-node cluster

[Fabio M. Di Nitto]

On 07/02/2012 11:39 PM, urgrue wrote:
> On 2/7/12 19:14, Digimer wrote:
>> On 07/02/2012 01:08 PM, urgrue wrote:
>>> I'm trying to set up a 3-node cluster with clvm. Problem is, one node
>>> can't access the storage, and I'm getting:
>>> Error locking on node node3: Volume group for uuid not found: <snip>
>>> whenever I try to activate the LVs on one of the working nodes.
>>>
>>> This can't be "by design", can it?
>>
>> Does pvscan show the right device? Are all nodes in the cluster? What
>> does 'cman_tool status' and 'dlm_tool ls' show?
>>
> 
> Sorry, I realize now I was misleading, let me clarify:
> The third node cannot access the storage, this is by design. I have
> three datacenters but only two have access to the active storage. The
> third datacenter only has an async copy, and will only activate
> (manually) in case of a massive disaster (failure of both the other
> datacenters).
> So I deliberately have a failover domain with only node1 and node2.
> node3's function is to provide quorum, but also be able to be activated
> (manually is fine) in case of a massive disaster.
> In other words node3 is part of the cluster, but it can't see the
> storage during normal operation.
> Looking at it another way, it's kind of as if we had a 3-node cluster
> where one node had an HBA failure but is otherwise working. Surely node1
> and node2 should be able to continue running the services?
> So my question is, do I have an error somehwere, or is clvm really
> actually not able to function without all nodes being active and able to
> access storage?

CLVM requires a consistent view of the storage from all nodes in the
cluster. This is by design.

A storage failure during operations (aka you start with all nodes able
to access the storage and then downgrade) is handle correctly.

Fabio