Backups are automatically-created disk images of Droplets. Enabling backups for Droplets enables system-level backups at weekly intervals, which provides a way to revert to an older state or create new Droplets.
Backups are not encrypted at rest, but they are not externally accessible.
DigitalOcean’s automated backups do not include volumes, but you can take manual backups with volume snapshots.
DigitalOcean backups are optimized for systems with low write activity. For systems with active database writes and other high I/O workloads, application-level backup may be more appropriate.
The snapshots that are taken to create the point-in-time data set use a copy-on-write mechanism. Copy-on-write allows for instant snapshots which make them a good choice for data consistency. There is almost no overhead for the actual creation of the snapshot that will be backed up.
However, copy-on-write implementations do take a performance hit on new writes that occur after the snapshot is taken until the backup process is complete. This happens because, for every new write, a system using copy-on-write must read the original data, write it to a new location, and then write the new changes to the original data location. This can significantly impact performance on busy, I/O bound workloads. The performance impact disappears when the snapshot is automatically deleted after the backup operation.
Databases are especially affected by this. Most database operations are heavily reliant on disk I/O, which can cause the application or the backup process to bog down and sometimes fail. In addition to the performance impact, operations that reside in memory or cache that has not been flushed to disk will be lost. Crash-consistent backups always save what is on the disk, but never what is in memory or cache.
If you are running a database or another application that produces high I/O load, choosing an application-level backup method may be a better alternative, as many databases have backup solutions designed specifically to work with those systems.