[Linux-cluster] Multipathing, CLVM and GFS

Sajesh Singh ssingh at amnh.org
Mon Jul 28 14:21:46 UTC 2008



Christine Caulfield wrote:
> Sajesh Singh wrote:
>>
>>
>> Christine Caulfield wrote:
>>> Sajesh Singh wrote:
>>>> Centos 4.6
>>>> Cluster Suite
>>>>
>>>> I am currently running a 2 node GFS cluster. The storage is 
>>>> provided via a fiber channel connection to the SAN. Each node 
>>>> currently has  a single FC connection to the SAN. I would like to 
>>>> migrate to using dm-multipath with each node having dual fiber 
>>>> channel connections to the SAN. Can I assume that CLVM is aware of 
>>>> the /dev/dm-# devices that are used to access the multipathed 
>>>> devices? Are there any gotchas that are associated with installing 
>>>> the device-mapper-multipath software after the GFS cluster is up 
>>>> and running? Are there any howtos available for this type of setup?
>>>>
>>>
>>> clvmd works fine with dm-multipath devices. You will probably have 
>>> to edit /etc/lvm/lvm.conf to exclude the underlying /dev/sd devices 
>>> to stop it getting confused though.
>>>
>>> You won't be able to do this with GFS mounted on the local node 
>>> though, you'll have to umount it, setup dm-multipath, vgscan & 
>>> remount. You CAN leave them mounted on other nodes while you do it.
>>>
>> Christine,
>>             Should clvmd be restarted as well so that I can create 
>> new volume groups? I have device-mapper-multipath setup, but if I try 
>> to run pvcreate /dev/mapper/mpath1p1 the command just hangs without 
>> any errors.
>
> No, you shouldn't need to restart clvmd unless you have an old version 
> and have added or removed PVs (which might be the case if you have 
> changed multipath devices).
>
> pvcreate does very little in the way of cluster operations so it's 
> probably worth checking it's not stuck writing to the dm device.
>
> (sorry for the delay in replying, I've been away).
>
>
Thanks for the reply. All is working now. I did add PVs, but did not 
need to restart as it might just have been the clvmd cache. All was okay 
after refreshing the cache.






More information about the Linux-cluster mailing list