Avamar - Exchange VSS backup fails on dismounted database

I've noticed something very similar.  When a backup runs, Avamar makes one request to VSS to create a snapshot of all of the volumes which hold all of the data and log files needed to backup all of the databases in the Avamar Exchange dataset.  If any part of the snapshot attempt fails, Avamar fails the entire backup job.  Thus, if the Exchange databases in a single Avamar dataset include more than 64 volumes (the maximum number of volumes for a single VSS shadow set), the backup will fail.  With Exchange 2010, it's pretty easy to have 30-plus databases on a single Exchange mailbox server in a medium sized environment.  If each DB has one data volume and one log volume, you'll hit the 64 volume limit rather easily.  As you've seen, the VSS snapshot attempt will also fail if the Exchange databases aren't in a healthy state (so as to not allow you the false sense of security provided by a backup of potentially bad data since Exchange can't know what has happened to the data since the DB was dismounted).  The workaround, as it sounds like you've discovered, is to manually create datasets that don't violate any of the VSS constraints (e.g. number of volumes, offline databases, etc.)  A better solution, in my opinion, would be for Avamar to loop through the Exchange databases to be backed up and request a seperate VSS snapshot per Exchange database and only fail the backup of the particular databases for which the VSS snapshot request fails.  This would allow the backup job to complete with exceptions, which would be similar to how Avamar handles SQL backups today.  I swear I've submitted an RFE before with this suggestion but I cannot find any confirmation at the moment of having done so.  This would be an excellent enhancement in my opinion.  As another work around, Avamar support may be able to supply a flag you can add to the Avamar Exchange dataset to force Avamar to ignore dismounted databases.  I don't know if such a flag exists but it sounds like something that would be possible to implement and it would be useful.  We switched to manually defined datasets for Exchange backups and haven't had too many issues since (other than having to manually change the datasets as Exchange databases come and go).