vCenter and Associated SQL Failover

[ekisner]

Hi there, this is partially a question, and partially a dump of my thoughts, so that other peoples thoughts can be added.

My vCenter server, as well as the MS SQL server that the database is hosted on (separate VMs) are replicated to our DR site. I have in the past had trouble with failing over my vCenter and the database.

The problem, which makes perfect sense, is that when the vCenter vm is powered down for a graceful failover, obviously the service is unavailable. Or, when the SQL server is powered down for a graceful failover, it too becomes unavailable, which in turn kills the vCenter.

Some solutions that come to mind, most of which I've found in this very forum (and I'm sure those I didn't find here are in fact just ones I didn't see):
1) Linked vCenter.. this would overcome the issue of trying to replicate the production vCenter server - largely because it would not be replicated. It has a cost licensing cost however, as a simple cold standby apparently isn't really a good idea as of vCenter 4+. Does not address the issue of the database.
2) SQL Mirroring would overcome the issue of trying to replicate the SQL server. Soon as the one went down for failover, the other would be available. Could maintain the mirror at the DR site for improved failover times. Could even go a step further and set up MS SQL Express on vCenter to act as the witness
3) Replicate vCenter, Mirror SQL - vCenter is mostly stateless. Anything important is in the DB and anything unimportant is well, unimportant. Do an unplanned failover of the vCenter server (ideally right after a manual replication pass?), and then after power-up is complete proceed with failover as planned for everything else, since the mirrored SQL server will handle the potential for vCenter outages. I'm never in favor of yanking the (virtual) power from a production service... but as far as this one goes, pretty minor impact? Would VSS potentially reduce this impact?

Anyone have any thoughts? I'll probably implement #3, as I can't really justify purchasing a vCenter license for my DR site (especially since I have more than 3 hosts at that site).

[Vitaliy S.]

Hello,

In addition to what you've just said, you may also want to consider following these options:

1. Replicate your vCenter Server and SQL Server as you're doing now and then perform failover operation via vSphere Client directly. In this case you would just power on your SQL Server and vCenter Server VMs at the remote site and that's it.
2. Another way would be to add your DR hosts to the backup console as standalone (via IP address or hostname) and then initiate the failover operation via backup console.

P.S. as far as I know under rare circumstances there could be some issues with the second option, but it's going to be addressed in the upcoming version.

Thanks!

[ekisner]

Interesting... I hadn't even thought of just adding the DR site hosts in without a vCenter server. That would probably be the easiest.

I'm not sure I'd ever have the guts to try the first option though... kicking an SQL server in the pants and then asking it to boot up nicely is asking for trouble. Well... that and it hosts other databases as well, so the risk isn't something as small as just the vCenter server.

But your second option... yes... makes me very glad that I put this post here. Very excited for B&R 7!

[Vitaliy S.]

kicking an SQL server in the pants and then asking it to boot up nicely is asking for trouble.

Actually there is no much difference (in terms of power on operation) whether to kick in the failover operation via Veeam backup console or start your SQL Server directly from the vSphere Client. If you do that via Veeam backup console you will be able to do RE-IP and then failback to the production site, but the power on operation is the same in both cases.

[ekisner]

Actually there is no much difference (in terms of power on operation) whether to kick in the failover operation via Veeam backup console or start your SQL Server directly from the vSphere Client.

Ah, yes, that's because we were talking about similar but dissimilar things. I absolutely agree that an actual failover operation would be just the same. I was referring specifically to the case of a graceful failover (ie in the event of a known impending hardware outage) being preferrable to an actual failover. During an actual failover there's not much choice . Apples and oranges, but that's where my perspective was coming from.

Question for you... will there be some form of application layer clustering for Veeam B&R in the new version?

[Vitaliy S.]

No, that's not going to be available.

[ekisner]

Darn. Ah well, thanks for your time Have a great day.