Backing up distributed application

[Moebius]

Hello,

I have some cases of an application that is distributed on several VMs. For example, there is a CAD application that has the Oracle database on a VM, and its data files on another one. The database keeps track of the links among the dozens of data files related to each CAD object.

The DB is changing all the time, reflecting all the changes to the data files (additions and deletions). You can llok at it as a database having its blob objects on the file system of a different VM.
In this case, the two VMs should be backed up at once, since it is essential that the snapshot of VM #1 is taken at the same time as the snapshot of VM #2. Otherwise, if the snapshots were taken at different times, the database content would not be coherent with the data files.

I have tried to figure out how to synchronize the snapshots that occur prior to the backup, but couldn't come up with any idea. Anybody knows a solution to this problem?

Thanks,
Lucio

[tsightler]

There's no real way to do this currently with VMware. That being said, you should be able to get very close by having each VM in a job and starting them at the same time.

That being said, certainly the application has some documented method for recovery as there's really no backup solution that can do this, except for perhaps hardware based SAN snapshots, which are not really a backup. Most likely they rely on Oracle's "point-in-time" recovery to be able to recover the database to a specific point-in-time using archive logs. You can easily design a strategy that leverages Veeam along with Oracle recovery (either traditional media recovery, or ideally RMAN). The recovery scenario would require two phases, Veeam would recover the two VM's quickly, and then you would leverage RMAN (or Oracle media recovery) to recover the database to the same point in time as the application backup.

That being said, I've worked with similar systems before, and in most cases it's fairly easy to manually recover from a slight "out-of-sync" scenario. Typically objects that are checked in/out during the time difference are the only things actually affected, and can normally be recovered fairly trivially, especially if you run the backups during a "quiet" time when large amounts of such activity are likely to be low.