how to roll back a repository transaction

[geek-andi]

I misunderstood recreate AND entered an invalid recreate command. Can I recover the data from borg-test::initial? borg-test still has its original size on the filesystem. (the borg-test repo ist just for demonstration what happened)

borg 1.4.1

~$ borg create borg-test::initial test-data --stats
------------------------------------------------------------------------------
Repository: /home/andi/borg-test
Archive name: initial
Archive fingerprint: 5cb7e81700c428e71e00ada0168c6101ed6ec560f3ee133cdcab184b3acf5653
Time (start): Thu, 2026-01-22 22:51:49
Time (end):   Thu, 2026-01-22 22:51:49
Duration: 0.00 seconds
Number of files: 1
Utilization of max. archive size: 0%
------------------------------------------------------------------------------
                       Original size      Compressed size    Deduplicated size
This archive:                  524 B                494 B                494 B
All archives:                   23 B                 26 B                687 B

                       Unique chunks         Total chunks
Chunk index:                       3                    3
------------------------------------------------------------------------------
~$ borg list borg-test
initial                              Thu, 2026-01-22 22:51:49 [5cb7e81700c428e71e00ada0168c6101ed6ec560f3ee133cdcab184b3acf5653]
~$ borg list borg-test::initial
drwxrwxr-x andi   andi          0 Thu, 2026-01-22 22:51:40 test-data
-rw-rw-r-- andi   andi         23 Thu, 2026-01-22 22:50:43 test-data/test-file
~$ borg recreate borg-test /this/does/not/exist::initial borg-test
Processing initial
~$ borg list borg-test::initial
~$ 

[ThomasWaldmann]

Yeah, recreate used wrong can be rather destructive: looks like you removed all files from the archive(s) by filtering with a non-existing path.

IIRC we have a warning in the docs addressing that and how to avoid it.

In any case, make a backup of the repository directory before doing anything else and do NOT run borg compact!

I you look into borg-test/data/<highest_number>/... you will see a bunch of segment files, these have just an increasing number as file name and they are grouped into respective directories (also numbered).

The last file you see is the N + 2 file, with the commit.

n .. N: The big files have new data chunks (generated by borg create or other borg ops) in them (PUT tags + id + data), maybe also DEL tags + id to mark chunks as deleted. Many of them are 500MB size, except the final one, which has the remainder.

N + 1: The small file (usually a few kB, depending on the amount of archives in the repo) is the repo MANIFEST, borg writes it to a separate file, right before the final commit.

N + 2: The tiny file (iirc 17b long) is a COMMIT tag, borg writes it to a separate file usually. This marks the successful end of a transaction. Each borg operation creates a transaction in the repository.

If you go backwards in the numbered files, you will find the commit of the previous borg transaction (also 17b).

If you remove all files after the commit of the previous transaction, you have basically undone the transaction (which corresponds to the last borg operation that modified the repository).

You need to delete the cache afterwards, so borg builds it freshly from the repo data: borg delete --cache-only REPO

Then use borg list to check if the repo is in the desired state.

You can also roll back multiple transactions that way by repeating this process.

[ThomasWaldmann]

Feature idea: for borg 1.x, we could have a borg undo (or borg rollback) command that automates this.

Problem: it would only work ok, if borg compact has not been used after the transactions that should be rolled back.

[geek-andi]

@ThomasWaldmann Wow, thanks for your response!

What I did:

• Backup
• Delete all files in borg-test/data/<highest_number>/... after the initial backup as I did some more actions (timestamp helped identify all actions after the initial backup)
• borg delete --cache-only borg-test
• borg list borg-test
  • => Cache is newer than repository - do you have multiple, independently updated repos with same ID?
• Moved security dir as suggested in https://docs.borgbase.com/faq/#repair-damaged-repositories
• borg list borg-test => OK
• borg extract borg-test => Successfull

I was wondering, if borg should run the command in the first place if it receives an invalid path?

[ThomasWaldmann]

@geek-andi borg can't know beforehands whether a path is "invalid".

Even after sequentially processing all the source archive, it could only know if the path matched something or not, but not really know what you intended to do.

That's why you first need to run with --list --dry-run and decide yourself from the output if that is what you wanted, as our docs suggest.

[geek-andi]

it could only know if the path matched something or not

@ThomasWaldmann Sure, that's what I meant!

Thanks for letting me know! Yes the docs are clear on the risks, I was just not fully awake ;)