SRM+SQL+3rd party replication and quiescing the DB

[TedH256]

Gentlemen 

What can SRM customers to ensure that a SQL VM running on (and replicated by) 3^rd party storage is application-consistent? Does the HP SRA have code that can invoke quiescing within the VM? Or is there some sort of scripted way, within the context of SRM, that a DB quiesce in the VM can be triggered in tandem with or just as a LUN replication occurs?

I know that HP is responsible for the SRA – but in these situations we get sometimes caught between the vendors. Looking for some “how have others protected mission critical SQL resources using 3PAR and SRM” experience.

Said another way:

We are looking for assistance with a DR implementation for a specific SQL application. The app is on VMs running on 3PAR storage. We have purchased all the required “3PAR application suite licenses” and then also have VMware SRM.

We are most concerned that LUNs replicated from production to DR site (LUNs will be datastores hosting the SQL VMs I presume) will be fully SQL-consistent (i.e. better than crash-consistent), when replicated.

We have some experience with 3PAR replications – but to my knowledge have not yet worked with the application suites.

Question – can anyone describe what role (if any) the 3PAR application suites may have in properly quiescing SQL running on a VM that is on a 3PAR LUN? I for one am not sure it can be done this way – seems that we may need to use RDMs (or vVOls if they are SRM compatible) to properly quiesce the DB. Is there an HP storage agent that can run on a SQL VM to coordinate with the array when snapshots of the underlying LUN are executed?

[vbrowncoat · Accepted solution]

You are correct. Application consistency would have to be handled at the array level. Some array vendors (I know of NetApp specifically) provide the ability to place an agent on VMs and have that agent work with the array to ensure application consistency, I'm not sure about HP.

Another option you could look at is vSphere Replication. VR offers VSS quiecing (along with MPIT) natively.

If you're looking for a specific answer to how to use 3PAR application replication suites I'd suggest posting to an HP forum. vVOLs aren't compatible with SRM yet and I don't see how RDMs would get you what you are looking for.

Lastly, to how you addressed your post, I would suggest that you not assume that only men participate in this forum.

[TedH256]

I'll take the redirection about the gender of this audience - agreed, was meant "globally" but should have used a different greeting.

Anyway - I think you are correct on the other bits as well - I have just found the following document, which I believe details the process to be used: http://h20195.www2.hp.com/V2/GetPDF.aspx/4AA5-0979ENW.pdf

Essentially:

replicate the LUN with the SQL VM OS drive, and include it in the SRM recovery plan. However replicate the LUN that holds the SQL data separately and do NOT include in the recovery plan. The SQL DB drive would be an RDM (so that it could make use of the 3PAR application suite quiescing of SQL). Then define a "post recovery" process for recovering the SQL DB after the VM is recovered.

I am curious - I was not aware that NetApp had an agent or whatever that could be installed on a VM - I thought the snapdrive tools were just for physical servers. But maybe if the drive is an RDM? And then still there would be the question of how to coordinate the quiescing with the replication intervals - is that something that the SRA would coordinate?

Also - with vSphere replication - is the tools provided VSS quiescing fully SQL aware? I believe that a couple years ago anyway, the best that could be achieved with tools was "crash consistency"?

[vbrowncoat]

Check out this KB for details on VSS - OS vs application level. It depends on OS and app but in at least some cases it supports app level with SQL.

Regarding NetApp application scheduling, from this doc: http://www.netapp.com/us/media/tr-3822.pdf I found it looks similar to what you mentioned, using a RDM. The SRA wouldn't coordinate it, it would just be part of the replication configuration on the array.