vMotion and guest clustering

[canz]

Hi All

Today we clustered our esx servers and we can migrate machines between esx hosts as expected

The problem as you might have guessed is that our OES2 (and i assume MSCS or any other cluster guest) is now broken because we are unable to map the required luns to guests that physically live on our different esx hosts (i.e esx does not allow a lun to be maped twice)

I.E.

ESX1 = CLUSTER SERVER1   -> lun 1

   \-->  = WEB SERVER

ESX2 = CLUSTER SERVER2   -> lun 1 (can not be done with cluster enabled esc)

physical = CLUSTER SERVER 3  -> lun 1

so the web sever will go between the esx fine but the cluster no longer works correctly

Has anyone else found a solution to this,

ideally id like to say that CLUSTER1 & 2 can not fail between esx hosts and hence should be able to see LUN1

but there does not seem to be a way to bring the guest out of the cluster

or is there a setting that i can set for the guest that will show the LUNs rather that hide them

if you need more info let me know

cheers

Steve Canz

[a_p_]

I'm not sure whether I understand the issue correctly.

There are some limitations for MSCS (http://kb.vmware.com/kb/1037959) including vMotion/DRS being not supported for VM's with shared disks.

However you should be able to cold migrate VM's between the hosts (depending on the setup, i.e. Cluster-in-a-box or Cluster-across-boxes) if all the hosts have access to the RAW LUN as well as the RAW device mapping file (vmdk) which was created when you first mapped the LUN as a RAW device.

If you want ot keep the MSCS nodes on dedicated hosts, you can configure (host) affinity rules (MUST rules). Depending on the vSphere version, see http://kb.vmware.com/kb/1004617

André

[canz]

Thanks for the advice

I thought affinity rules might be the answer but our vMotion licen ce does notcover DRS ( only HA )

just to summarize the problem

i have two esx servers with a cluster server on each, these cluster machines will never move from there host.

i have a lun that both cluster guests can see

after mapping the lun to one guest the other guest can no longer see the lun

thanks again

Steve Canz

[idle-jam]

yes you're right ..

[a_p_ · Accepted solution]

... after mapping the lun to one guest the other guest can no longer see the lun

This is by design. You can attach a LUN as a RDM only once. On the second node, you have to attach this RDM-LUN like a virtual disk, selecting the mapping vmdk file which was created when you first mapped the LUN. To be able to do this, you need at least another shared LUN with a VMFS datastore, on which the mapping file has to be located.

André

[canz]

thankyou, this is the answer

now to start another post for my next question

How can i attach a 2.5TB LUN

File [CL_VM_Guests] server/server.vmdk is larger than the maximum size supported by datastore CL_VM_Guests

[mcowger]

You can't.  If you need > 2.0TB you need to split it up into 2 LUN of < 2TB each or use an in-guest iSCSI initiator to access the LUN.

[a_p_]

In addition to the max. LUN size limit mcowger already mentioned, you need to select the appropriate block size when creating/formatting a new VMFS datastore. For block sizes and the corresponding max. file sizes see e.g. http://kb.vmware.com/kb/1012384

Since there are no performance issues with a larger block size, you may consider to create your datastore with an 8MB block size to support virtual disks with up to the limit of 2TB minus 512 bytes.

André

[canz]

Thankyou all for your help

An update on the cluster issue

because vmware locks the vmdk file when a guest is loaded ive been forced to move the Cluster1Server folder ( that contains the vmdk lun links ) to Cluster2Server then delete and recreate the lun's from Cluster Server 1

now i can map the luns to ClusterServer2 using the moved vmdk files, I know this isn't supported but it seem the only solution

as far as the 2TB limit, this kinda suk's so ill make multiple 2TB volumes and hope vmware decide to increase there block size in the future

thanks agian