Change backup process to stream to disk

We have ~1G of attachments.

Simultaneously migrated:

• Confluence from 2.2.5 to 2.2.9
• the attachments from the file system to the DB
• JVM from 256mb max to 512mb max.

The backups now throw an OutOfMemory exception. Even if I de-select the option "Backup Attachments" Confluence still attempts to load the entire recordset into Hibernate memory, resulting in the exception.

I understand that we can use a MySQL backup and a replica of the Confluence Home directory, however an integrated backup would be nicer for our system – and even an automated backup with attachments would be very nice to have. We have been stung in the past with using this restore process failing (since appropriate changes to the database schema are not made when you restore direct to the DB instead of using Confluence).

I propose a change to the backup process to use a streamed-out ZIP file, or at the least not requiring full load into RAM of the entire JVM recordset (instead of doing an eager load of the db, just load one record at a time ... )