MSCS and DRS HA

[luizhbedin]

Hi all,

We are designing a big transformation project adn we are facing a issue with Windows clustered servers, we know that vmware doesn't recommend insert clustered machines to a DRS / HA cluster, but one question remain in my mind: Can I put the clustered VM into the clustered environment and just disable the VM to be a part of HA and DRS ?

Thanks

[BryanMcC · Accepted solution]

You would have to leave the host out of the HA cluster in this case. Now you may get away with not allowing the VM to be powered on in case of failure but that would defeat the pupose. I have not tested any of the MSCS with HA / DRS but from what I understand it does not work. But I did find this link in which I have to agree with the author may not be suitable for production in which support may need to be consulted.

Happy clustering

Help me help you by scoring points...

[gmjulian]

Do you mean that you want to disable the clustering service on the OS once you move it to VMWare? If so, you will render the server useless. A cluster server keeps all of its resources grouped by resource groups through the cluster administrator. If you disable the cluster service, you have no resources. If you are willing to recreate and verify the resources after you evict the nodes from the cluster (essentially uninstalling clustering) you could do it that way.

Keep in mind a few things about MSCS & HA

1) MSCS servers on vmware esx hosts need to have there system disk residing on a local VMFS volumes. This means no V-motion as you would never want the 2 servers to sit on the same host ( defeats the purpose)

2) HA really provides similar advantages as to clustering 2 or more physical boxes. They allow you to overcome an outage in case of a physical hardware failure by moving that session to the remaining nodes (or esx hosts ) in the cluster. However, with HA , the machine goes down hard. So if you have an app or database that is sensitive to sudden shutdowns, you may experience file or database corruption. Make sure you are dumping transaction logs frequently if this is the case and you go the VMware route.

3) The one advantage of physical clustering is that you can have a host go down or corrupt the OS for whatever reason and you have another OS ready to take on the responsibility of what was running on the failed node without the data going down hard. With running a single OS on a VMware env, if you somehow corrupt that session you are restoring from a snapshot or the vmdk file(s).

[mreferre]

1) Very true

2) True but after all ...... many MSCS implementation are set up just to protect from a node hardware outage so the service would go down hard anyway in both cases..... It is true that an MSCS clustering can have more granularity in terms of resource monitoring but at the end of the day this is rarely leveraged/used (my opinion).

3) Very true. One of the problems with HA and the single vm scenario is that if something goes wrong with that single OS image (ntfs corruption, OS upgrade issue etc etc) you are toasted. In a MSCS cluster you can use/do rolling upgrades whereas if something goes wrong on the first node you still have the other node. In reality though MSCS is a strange beast and is not uncommon to find customers that say that MSCS brought in more troubles than those that it has solved. Perhaps an approach comprised of HA + diligent backup of the vmdk / data is a better approach.

I can't say to be a fan of MSCS on VI3. In my opinion either don't use MSCS on Vi3 or if you have to use it (for technical or political reasons) consider running it on physical. This would be my general rule of thumb.... of course there might be niche situation where it would be advisable to use both together.

Massimo.

[Texiwill]

Hello,

HA & DRS does not affect MSCS because the C: drives are all on Local storage (per the MSCS documentation Page 22 of http://www.vmware.com/pdf/vi3_301_201_mscs.pdf). You can not vMotion from local storage or access local storage from a remote site. So it is possible to put MSCS or any shared disk cluster within an ESX cluster and HA/DRS will not affect them in any way. However to be sure just exclude them from 'auto' power on.

Share disk clusters, shared across multiple ESX servers or with physical hardware is just one way to handle failure modes. I see their use and sometimes use them with customers and other times not. In an HA situation, the VM is dead anyways and the VM will automatically be booted on a new host. The guest is booted using a crash consistent virtual disk, which means it may not come up at all or if it does may be missing data. In a shared disk cluster situation, the other nodes of the cluster (up to 8 for MSCS within ESX) are supposed to take over the load, it takes about the same amount of time for a clean reboot as it does for the failover in some case, yet you do not necessarily need to worry about crash consistency, but there may be a little missing data.

They both are there to solve different problems actually, and I am not sure either can be replaced with the other.

Best regards,

Edward L. Haletky, author of the forthcoming 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', publishing January 2008, (c) 2008 Pearson Education. Available on Rough Cuts at http://safari.informit.com/9780132302074