SharePoint and SQL Snapshot backups and Microsoft Support

[PNWMtnBiker]

Our SharePoint admin just stopped by and told me that his Microsoft support rep informed him they will not support a virtual SharePoint farm that is being backed up by snapshots such as how Veeam does it and same goes for SQL. And then how we should buy and use data protector instead.

What do you guys have to say about that? This was certainly news to me but can't be the first time this has come up.

*edit, I've done some quick reading up on SharePoint Farm snapshots for backup and the problem comes from performing a restore and the timer service getting out of sync with the other servers. The articles say it shouldn't ever be done without powering down the whole farm and the SQL server entirely and making the backup. Then the entire farm and SQL server have to be restored together if a restore is needed.

[Gostev]

First time I hear. We have over 100'000 customers backing up SQL and SharePoint, some for over 6 years now, and they have never encountered any issues with Microsoft support (otherwise, I would know). Most likely, this is just an opinion of specific support rep, because Microsoft's own backup product (DPM) backs up SharePoint and SQL Hyper-V VMs using snapshots. Please ask him/her to provide the link to official statement, and we will work with Microsoft to clarify this. Thanks!

[PNWMtnBiker]

Thanks Anton,

The SP admin gave me this link below to an msdn blog post. I've asked him for the contact info of the MS rep he spoke with so I can see about some official MS documentation.

http://blogs.msdn.com/b/mossbiz/archive ... shots.aspx

[Gostev]

This blog post is correct, but it does not apply to Veeam. The author is talking about implications of using regular VM snapshots on SharePoint. These issues are well known since the beginning of virtualization, and are definitely not specific to SharePoint. The most classical example so far has been Active Directory, as by now VM snapshots have ruined unheard amount of domain controllers all over this planet.

Essentially, there are two types of issues:

1. Most issues (and most evil issues) with applications happen due to people creating, and after some time reverting back to the created snapshot. This article provides three different examples on when doing this will break SharePoint (issues #1, #2 and #4). Veeam never reverts back to snapshot created during backup, instead the snapshot is simply discarded after backup.

2. Another type of issues happen because with regular VM snapshots, applications are unaware that they are being "backed up". One example of consequences is issue #3 (quite minor tbh), however there are much more severe consequences in case of Active Directory and Exchange, where this causes not just data loss, but "production down" situation. To avoid these kind of issues, Veeam integrates with Microsoft VSS to prepare applications for hot backup, and to let them know they are being backed up.

Hope this helps!

[PNWMtnBiker]

Thank you very much for the reply Anton, this helps clarify things greatly.

I have a follow up question if you don't mind, is there some documentation from Veeam that explains on a deeper level this; "Veeam integrates with Microsoft VSS to prepare applications for hot backup, and to let them know they are being backed up."

I have an understanding of how VSS works, but I want to share this with our SP admin as he is specifically asking for something.

[PNWMtnBiker]

One more question

Does it matter if the farm is queisced with VSS at the same moment or not? Of course I intend to have the whole sharepoint farm in the same job so I know it will be really close but if one server starts and say the db server start 45 seconds or a couple minutes later, is there any impact or concern if I had to do an entire farm restore later on?

And our sysadm also asked me to thank you Anton for reply about the article, it is much appreciated

[Gostev]

Q1.

Well, sticky FAQ topic has higher-level explanation of our application-aware image level backup technology including VSS integration which might be sufficient for your SP admin, now of course User Guide goes into more details on this topic, but let me give you my own 10 lines tech pitch.

In short, we integrate with VSS "by the book" of Microsoft guidelines as any "classic" backup application is supposed to do, but we intercept the moment when the OS is ready to create VSS snapshot, and issue VM snapshot creation command at this very moment, so the resulting VM snapshot contains application-consistent and quiesced server state. We then still let VSS snapshot happen normally as well, so the OS does not mention our "intervention" at all. We have to take this creative and unique approach because VSS was not designed with image-level backup in mind, nevertheless our approach is fully validated directly by Microsoft VSS development team, who at one point became so interested in what we are doing, that they requested us the full review of our integration architecture.

On top of basic VSS integration, which is generic and works for any VSS-aware application (application specifics is handled by per-application VSS writers produced by application developers and included with the application), our application-aware processing logic also includes detection of specific applications running in VM, and performing additional application-specific steps required for backup and image-level recovery of that specific application. Some examples include preparing the application for recovery on the first boot up of the restored VM image, truncating Exchange logs on successful backup, detecting SharePoint server to SQL server mappings to enable granular recovery, and so on.

Q2.

No, it does not matter. Besides, not a single backup solution in the world is able to deliver quiescence across multiple servers at the same moment anyway, no matter what backup technique they are using (file-level or image-level). Pulling something like this across multiple servers is simply impossible even in theory.

[gingerdazza]

Hi Gostev, an old post I know, but wouldn't Zerto be able to give you consistency of a multi-VM application stack (i.e. SharePoint)? It can use VPGs to contain multiple VMs, and transmits all hypervisor I/O for the entire group of VMs with write order fidelity. This atop of quite long PIT rollback capabilities would protect the entire stack with full application consistency, no?