We understand your concern, and treat reliability and security very seriously.
We store source code repositories on enterprise class storage using RAID 6 that is designed to be highly available. Our compute nodes are stateless, load balanced, and redundant. We take daily snapshots of your data. We retain these snapshots on our primary systems, replicate them to an offsite data center, and maintain them in both locations for seven days.
Our backups are for disaster recovery purposes only. We are not able to use our backups to restore repositories that have been deleted by an end-user.
Our database, which stores information such as accounts, issues, and comments, is replicated to separate hardware and we perform SQL archives every four hours. We distribute these database backups between two servers and write them to tape daily, which is then taken offsite. Each backup tape is maintained offsite for at least 30 days.
One of the great features of distributed version control systems (DVCS) such as Git and Mercurial is that they're distributed. Backups are still very important, but consider that every person in your workspace that has forked your repository has a complete copy with a full history. Also, it's very easy to setup mirrors and automated backups to pull recent changesets. This is something the Atlassian Community can help make recommendations on.
Note: On rare occasions, our Support team can help recover core repository content only, as long as the deletion happened within the last 7 days. We are unable to recover metadata or any content that was deleted past 7 days.