Yep; image based restores will tend to bone you if the VM is a replication 
partner with something else or if it has a database engine running on it. What 
people often tend to forget is that database engine doesn't just mean SQL, it 
also means JET/ESE as well. 

Self contained databases can be ok if it's just for an on box app (say 
WebJetAdmin) or the like. 
I've found the best use for them to be application servers with no on box data. 

DAMIEN SOLODOW
Systems Engineer
317.447.6033 (office)
317.447.6014 (fax)
HARRISON COLLEGE

-----Original Message-----
From: [email protected] [mailto:[email protected]] On 
Behalf Of Ben Scott
Sent: Thursday, September 12, 2013 10:39 AM
To: [email protected]
Subject: Re: [NTSysADM] In defense of image-based VM backups

On Wed, Sep 11, 2013 at 10:53 PM, Richard Stovall <[email protected]> wrote:
> There was a discussion here a few weeks ago that centered on 
> image-based backups for entire VMs vs data-based backups of 
> applications only.  My recollection is that most of us smaller guys 
> prefer (or at least lean
> toward) the image-based variety, and the bigger, more mature orgs 
> prefer data backups.

  As I understand it, image backups are actually not a problem at all.
 It's restores that cause issues.

  A program may use things like serial numbers, Update Sequence Numbers, etc., 
to track activity internally.  Say your SQL database is at USN 1002, and you 
take a snapshot of the SQL host.  Then you update a record to have a new phone 
number, and that change gets assigned USN 1003.  Then something bad happens, so 
you restore, and now we're back at USN 1002.  Someone else submits a delete of 
a different record, and now *that* gets USN 1003 again.  Meanwhile, some other 
part of the system/network that wasn't restored thinks USN 1003 was an update.
Weird behavior ensues.

> Long story short, having the ability to (almost) immediately spin up a 
> brand new, sandboxed copy of the CRM server allowed me to experiment 
> and figure out how to resolve the problem without touching the one the 
> devs actually use.

  In a sandbox scenario, the restore issue described above is not an issue.  
What you describe here is a tried and true technique.  Just make sure your 
sandbox is good.  :)

-- Ben




Reply via email to