dfb 20241231.0b0

Dated File Backup

---

Warning: This tool is still in beta but (a) I've been using it heavily for 1+ years without problems, (b) the design is such that even without it, you can get all of your files back, and (c) more testers are good! Use at your own risk but please also provide feedback

---

dfb - Dated File Backup

The dfb backup tool utilizes rclone to create full-file, append-only backups that allow easy restoration to any point in time. Files are uploaded with the upload date appended to its name. The design focuses on simplicity, easy understanding, and restoration without special tools. It prioritizes full-file backups stored natively on the remote, continuous rollback capability, straightforward backups and restores, and support for append-only storage. To accomplish this, it sacrifices some efficiency and advanced features. It may not be the the most efficient, advanced, fast, featurefull, sexy, or sophisticated backup tool but these are great tradeoffs for peace-of-mind with backups!

Many other popular backup tools/strategies including macOS Time Machine, rsnapshot and rsync with --link-dest (example) have the same (or even worse) tradeoffs and don't support cloud storage. Block-based backup tools like restic, kopia, borg, and Duplicacy offer increased efficiency and deduplication but are more prone to errors/corruption, more complicated, and are all but impossible to restore without the original tool. Their complexity also leads to increased risk of corruption.

Design Tenets:

• Easy to understand, interrogate, and restore. The backup format is easily comprehended and can be reverse engineered simply. No special tools are needed to restore in theory (except if using crypt, you need to decrypt it). The format is about as straightforward as possible!
• Backup full copies of all files. All files are full copies stored natively and can be downloaded right away (except for reference files), with or without special tools. This is less efficient than block-based tools, but that comes with the advantages noted above.
• Restore to any point in time. Can easily rollback to any point-in-time in the backup.
  • Continuous in time is better than snapshots. Many tools work off of synthetic snapshots. Snapshots enable pruning like "keep 1 snapshot per week" but that is a risky approach. What if you need some file that falls in that range? Or what if you don't know when you modified a file? Instead, dfb can roll back to any point-in-time continuously and look at all versions of a specific file (with or without the tool itself to faciliate). dfb can still be pruned of "prune files older than XYZ days" and/or "prune all but that last N versions".
• Support append-only/immutable storage natively. There is never a need to delete files except for pruning. Nothing ever gets renamed, deleted, or modified.

Command Help

See CLI Help

File names

When files are backed up, they are renamed to have the date of the backup in the name. Filenames are

<filename>.YYYYMMDDhhssmm<optional R or D>.ext

where the time is always in UTC (Z) time. When a file is modified at the source, it is copied to the remote in the above form. If it is deleted, it is marked with a tiny file with D. If a file is moved, a reference, R is created pointing to the original. If moves are not tracked, then a move will generate a new copy of the file.

There may also be empty directory markers named .dfbempty.<date> which are to hold the empty directory on the destination.

Directory names are unchanged.

Reference Files

The only exception to full-file backups are references. References write JSON data like[^refv1]:

{"ver": 2, "rel": "<RELATIVE path to referenced file>"}

[^refv1]: Older versions were just a single line with the absolute path to the reference. This implied version 1 was less flexible and made for some other issues.

Note that references are considered and guarded when pruning (with associated tests). Be careful when pruning manually so as to not break a reference file.

Install

Just install from github directly.

$ python -m pip install git+https://github.com/Jwink3101/dfb.git

Setup

To start, run:

$ dfb init path/to/config.py

The config file is heavily and extensively documented. It is read in Python without any sandboxing so make sure it is trusted! Some variables are defined inline including os and a few modules. The variables __file__ and __dir__ are pathlib.Path objects for the config file and the directory of the config file respectively. These can be used to specify paths to things like files for --filter-from rclone flag.

Simple Usage

Assuming this is a new setup, just run it:

$ dfb backup --config path/to/config.py

Or set the environment:

export DFB_CONFIG=path/to/config.py
$ dfb backup

Or, directly execute the config file (it has a custom shebang):

./config.py # <--- Assumed backup mode
./config.py backup
./config.py ls
# ...

Override

The config file can be overridden at the command line by specifying code to evaluate before and after the configuration file. In order to control if it is evaluated before or after the configuration file, the variables pre and post are defined. Consider the following example:

dfb backup --config config.py -o "
    if post:
        filter_flags.extend(('--filter','- *.new'))"

Local Database

A local sqlite3 database is kept that serves to greatly speed up interaction. The database is used for listing the remote either in the CLI or for use in backups. It is updated as real-time as is possible with uploads so even interrupted or failed backups should be correct (or, if not, it may re-backup but not lose anything).

However, if the remote is modified manually, the database should be refreshed. Or if you wish to interact with the backup on another machine, it should either be refreshed or copied.

Note that, depending on the configuration, not all data may be in the database. For example, if not using mtime for dst_compare, it won't be fetched. That will still be there for restore if the remote supports it.

The database is stored in:

<rclone cache dir>/DFB/<config_id from config>.db

Where "rclone cache dir" is found from: $ rclone config paths. See also Advanced Settings.

When refreshing the database, dfb will also optionally read the logs of all uploaded files to fill in source-specific metadata and hashes and speed up references. These are secondary to the operation of dfb and the native remote storage is sufficient to restore (up to the limitations of the remote).

Pruning

You can prune dfb by removing unneeded files at snapshots older than a set time. It will keep files older than the specified time if they are still the most recent or are referenced. It can also be told to keep a certain number of versions of specific files.

This pruning strategy is discting from snapshot-based backups where you keep a certain number of snapshots, but offers a more practical approach as it doesn't risk losing important data bewtween snapshots and makes it very easy to understand changes.

Pruning is as simple as shown in some examples:

$ dfb prune "365 days"
$ dfb prune 2023-08-11
$ dfb prune 2022-10-14 --keep-versions 10  # keeps 10 *additional* before the date
$ dfb prune now --keep-versions 20         # just keep the last 20

Files older than the specified time or verion number may still exist if they are the latest or are referenced.

Pruning individual files (Advanced)

Pruning individual files is supported through the advanced sub command. You need to know the "real-path" or "rpath" of the file. Assume $DFB_CONFIG_FILE is set for these examples.

For example, if you want to delete my/large/file.ext, you can do:

$ dfb versions my/large/file.ext --real-path --ref-count

You will see a table of all versions of the file. Note that --ref-count is included since if you delete a referent (Ref. Count > 1), it will make the referrer appear deleted (that may be the intended outcome!). Be careful about references!

Simply identify the versions you wish to delete, copy the Real Path, and run

$ dfb advanced prune-file my/large/file.<timestamp>.ext

You can, of course, do it all manually but (a) it won't check/delete references and (b) it won't refresh the database. To refresh after manual pruning, either call refresh or use the advanced dbimport command. See notes on formats for details.

Additional Docs

• (ADVANCED) Reading Backup and Prune --dump files
• Changelog
• CLI Help
• Using dfb with cold storage
• Compare Setting Guidance
• "FAQs"
• Mount (EXPERIMENTAL)
• Symlink Update

Known Issues

• Logging in dfb-mount is not complete nor does it make much sense at the moment.
• Even with logs of files using source data, restore always directly uses the remote. Future versions may offer a secondary method to restore that information for certain remotes.
  • Example: If the remote doesn't support mtime, even if mtime is used for backups based on the logs and local database, restore will not restore mtime.
• Deletes are marked with a 3 byte file (just ASCII encoded DEL). This is smaller than most block sizes so they take more space on a local filesystem.