borgmatic

How to deal with very large backups

Biggish data

Borg itself is great for efficiently de-duplicating data across successive backup archives, even when dealing with very large repositories. But you may find that while borgmatic's default mode of prune, compact, create, and check works well on small repositories, it's not so great on larger ones. That's because running the default pruning, compact, and consistency checks take a long time on large repositories.

A la carte actions

If you find yourself in this situation, you have some options. First, you can run borgmatic's prune, compact, create, or check actions separately. For instance, the following optional actions are available:

borgmatic prune
borgmatic compact
borgmatic create
borgmatic check

(No borgmatic prune, create, or check actions? Try the old-style --prune, --create, or --check. Or upgrade borgmatic!)

You can run with only one of these actions provided, or you can mix and match any number of them in a single borgmatic run. This supports approaches like skipping certain actions while running others. For instance, this skips prune and compact and only runs create and check:

borgmatic create check

Or, you can make backups with create on a frequent schedule (e.g. with borgmatic create called from one cron job), while only running expensive consistency checks with check on a much less frequent basis (e.g. with borgmatic check called from a separate cron job).

Consistency check configuration

Another option is to customize your consistency checks. The default consistency checks run both full-repository checks and per-archive checks within each repository no more than once a month.

But if you find that archive checks are too slow, for example, you can configure borgmatic to run repository checks only. Configure this in the consistency section of borgmatic configuration:

consistency:
checks:
- name: repository

(Prior to borgmatic 1.6.2, checks was a plain list of strings without the name: part.)

Here are the available checks from fastest to slowest:

See Borg's check documentation for more information.

Check frequency

New in version 1.6.2 You can optionally configure checks to run on a periodic basis rather than every time borgmatic runs checks. For instance:

consistency:
checks:
- name: repository
frequency: 2 weeks

This tells borgmatic to run this consistency check at most once every two weeks for a given repository. The frequency value is a number followed by a unit of time, e.g. "3 days", "1 week", "2 months", etc. The frequency defaults to "always", which means run this check every time checks run.

Unlike a real scheduler like cron, borgmatic only makes a best effort to run checks on the configured frequency. It compares that frequency with how long it's been since the last check for a given repository (as recorded in a file within ~/.borgmatic/checks). If it hasn't been long enough, the check is skipped. And you still have to run borgmatic check (or just borgmatic) in order for checks to run, even when a frequency is configured!

If you want to temporarily ignore your configured frequencies, you can invoke borgmatic check --force to run checks unconditionally.

Disabling checks

If that's still too slow, you can disable consistency checks entirely, either for a single repository or for all repositories.

Disabling all consistency checks looks like this:

consistency:
checks:
- name: disabled

Or, if you have multiple repositories in your borgmatic configuration file, you can keep running consistency checks, but only against a subset of the repositories:

consistency:
check_repositories:
- path/of/repository_to_check.borg

Finally, you can override your configuration file's consistency checks, and run particular checks via the command-line. For instance:

borgmatic check --only data --only extract

This is useful for running slow consistency checks on an infrequent basis, separate from your regular checks. It is still subject to any configured check frequencies unless the --force flag is used.

Troubleshooting

Broken pipe with remote repository

When running borgmatic on a large remote repository, you may receive errors like the following, particularly while "borg check" is validating backups for consistency:

    Write failed: Broken pipe
borg: Error: Connection closed by remote host

This error can be caused by an ssh timeout, which you can rectify by adding the following to the ~/.ssh/config file on the client:

    Host *
ServerAliveInterval 120

This should make the client keep the connection alive while validating backups.

Improve this documentation

Have an idea on how to make this documentation even better? Use our issue tracker to send your feedback!