On Wed, Feb 19, 2014 at 9:46 PM, Jay Ashworth <jra@baylink.com> wrote:
Why bother with a clustering FS, then, if you cannot actually /use it/ as one?
It is used as one. It is also a lot more convenient to have a shared filesystem, than a distributed volume manager. You could think of VMDK files on a VMFS volume as their alternative to a Clustered Linux LVM. Just because you have some sort of clustered volume manager, doesn't make your guest operating systems cluster-aware. With VMFS... if two guest operating systems try to open the same disk, hypothetically, the most likely reason would be that there is a split brain in your HA cluster, and two hosts are trying to startup the same VM. The locking restrictions are for your own protection. If the filesystem inside your virtual disks is not a clustered filesystem; two instances of a VM simultaneously mounting the same NTFS volume and writing some things, is an absolute disaster. Under normal circumstances, two applications should never be writing to the same file. This is true on clustered filesystems. This is true when running multiple applications on a single computer. There is such a thing as 'Shared disk mode': if your block target is Fibre Channel, and you are using Microsoft Cluster Services in VMs, but an extra virtual SCSI adapter on a VM in shared mode, is something that you have to explicitly configure (It's not the default). There are many good things to be said about having a single-purpose filesystem, which can be placed on a shared storage device --- which multiple hosts can mount simultaneously, and view the same file/folder structure, and touch resources corresponding to which applications are running on that node... That does not require cluster votes and majority nodeset Quorums to keep the cluster from totally going out, and it does not need questionable techniques such as STONITH ("Shoot the other node in the head") as a fencing strategy, for providing exclusive access to resources. Different hosts can access files corresponding to resources running on that host, and HA is able to fail virtual disks over, so the Highly-specialized filesystem achieves all the objectives, that it needs to be a clustered filesystem to solve.
- jra
-- -JH