Checking old backups
GBM AI reads your older backups back now and then, so damage on the drive is found before you need a restore, and a restore takes damaged files from the second copy.
On this page
A backup is read back once, right after it is made (Verify files after each run). Weeks later the same files can go bad where they lie: a worn sector on an old disk, a NAS that rewrote a block, a file someone opened and saved. GBM AI therefore checks older backups again from time to time, and tells you before a restore needs them.
What a check does#
For each backup it takes, a check:
- compares the backup's list of files (its manifest) with the finished marker the run wrote last;
- opens that list read-only and has the database check itself;
- reads its side files;
- reads every stored file back and compares it with its checksum.
A check never writes into a backup. It runs as its own entry in Activity, with a report like any run.
When checks run#
- How often: each job's Settings step has Check old backups: Every week (the default for drives and shares), Every month, or Off. For a cloud bucket the default is Off, because every check downloads the whole backup, which your provider may charge for.
- What is due: a backup is due once its last check (or the run that made it) is older than the period. The longest unchecked go first.
- When: while nothing else runs, the scheduled backups are not paused and the PC is not on battery; also right after a job's backup and second copy, while its drive is plugged in.
- Never in the way: a backup or a copy asked for during a check makes it stop where it is, and a check stops taking new backups after two hours. What it did not reach waits for the next one.
- Check now: the job's page has Check now, which checks every backup of the job.
When a check finds damage#
- The check's report says Damage found and lists each damaged backup: which files, and whether the second copy holds them whole.
- The job shows Damage found under Checks, Home asks for attention, and Genie's card offers Run a full backup and Check again.
- A backup found damaged is not copied to the second copy again.
- Run a full backup soon: the newest restore point is then whole again, whatever happened to the old one.
Restores and damage#
A restore works round damage by itself:
- A file whose stored copy is damaged is restored from the same backup on the second copy.
- A backup whose list of files cannot be read is read from the second copy's list.
- With no second copy, a point in time steps past a backup that cannot be read: the files it changed come back as they were in the backup before it.
The restore report says what came from the second copy, and which backup was stepped past.
A copy in a cloud bucket is not read from by restores in this version; a second copy on a drive or a share is.
Encrypted backups#
A backup encrypted with a password the job has since changed cannot be read with the new password. A check leaves it unchecked and says so ("encrypted with an earlier password of the job"); it is not called damaged.