Thomas WaldmannandClaude Fable 5.1 99ce853567 compact: build the chunk index once, not three times
"borg compact" built the chunk index up to three times per run and kept two
copies of it in memory at the same time:

1. get_repository_chunks() builds compact's own index (it needs the usage flags).
2. analyze_archives() reads the archive metadata objects, and resolving their pack
   locations lazily builds a second, identical index inside the Repository.
3. compact_packs() calls delete_chunkindex_from_repo(), which drops the repository's
   in-memory index, so the archive listing in cleanup_files_cache() built a third one.

The chunk index is the biggest structure borg keeps in memory (about 80 bytes per
chunk plus hash table overhead), so a large repository paid for that twice over.

Share compact's index with the repository while the repository reads through it:
get_repository_chunks() installs it as repository.chunks, so analyze_archives()
resolves pack locations through that same index. The repository only reads pack
locations and F_PENDING from an index, never the F_USED flags or the sizes compact
tracks for --stats, and stored index fragments zero the flags either way - so the
index the repository would have built is the very same one.

The sharing ends in compact_packs(): delete_chunkindex_from_repo() (#9748) drops the
repository's reference before the first store change, and compact does not hand it
back. Nothing needs it there - compact_pack() and merge_packs() get chunks=self.chunks,
--stats and save_chunk_index() use self.chunks directly. So while the persisted index
is gone, the repository holds no in-memory index that Repository.close() could persist
on an aborted run, and save_chunk_index() can empty the index in place with clear=True.

cleanup_files_cache() used to list the archives itself, after save_chunk_index() had
cleared the index, which is what built the third index. It now reuses the names
analyze_archives() already collected, so it needs no repository access at all - and
every archive's metadata is read once per compact run instead of twice.

Verified on the same repository: master and this produce byte-identical repositories,
identical persisted index/* fragments (completed, dry-run, no-op, interrupted and
slow-rebuild runs alike) and identical output, with 3 index builds before and 1 after.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Co-Authored-By: Claude Opus 5 <[email protected]>
2026-09-12 22:47:16 +02:00
2026-08-29 17:35:13 +02:00
2026-08-23 14:52:12 +02:00
2025-09-23 14:56:23 +02:00
2026-03-08 13:06:01 +01:00
2015-10-07 17:54:43 -04:00
2026-09-07 04:31:27 +00:00
2025-09-23 14:56:23 +02:00

This is borg2!
--------------

Please note that this is the README for (unstable) Borg2 / master branch.

For general project infos, versioned docs, screen casts, please see there:

https://www.borgbackup.org/

Borg2 is currently in beta testing and might get major and/or
breaking changes between beta releases (and there is no beta to
next-beta upgrade code, so you will have to delete and re-create repos).

Thus, **DO NOT USE BORG2 FOR YOUR PRODUCTION BACKUPS!** Please help with
testing it, but set it up *additionally* to your production backups.


What is BorgBackup?
-------------------

BorgBackup (short: Borg) is a deduplicating backup program.
Optionally, it supports compression and authenticated encryption.

The main goal of Borg is to provide an efficient and secure way to back up data.
The data deduplication technique used makes Borg suitable for daily backups
since only changes are stored.
The authenticated encryption technique makes it suitable for backups to targets not
fully trusted.

See the `installation manual`_ or, if you have already
downloaded Borg, ``docs/installation.rst`` to get started with Borg.

.. _installation manual: https://borgbackup.readthedocs.io/en/master/installation.html

Main features
~~~~~~~~~~~~~

**Space efficient storage**
  Deduplication based on content-defined chunking is used to reduce the number
  of bytes stored: each file is split into a number of variable length chunks
  and only chunks that have never been seen before are added to the repository.

  A chunk is considered duplicate if its id_hash value is identical.
  A cryptographically strong hash or MAC function is used as id_hash,
  e.g. (hmac-)sha256 or (keyed) blake3.

  To deduplicate, all the chunks in the same repository are considered, no
  matter whether they come from different machines, from previous backups,
  from the same backup or even from the same single file.

  Compared to other deduplication approaches, this method does NOT depend on:

  * file/directory names staying the same: So you can move your stuff around
    without killing the deduplication, even between machines sharing a repo.

  * complete files or time stamps staying the same: If a big file changes a
    little, only a few new chunks need to be stored - this is great for VMs or
    raw disks.

  * The absolute position of a data chunk inside a file: Stuff may get shifted
    and will still be found by the deduplication algorithm.

  Supported chunkers:

  * fastcdc (the default)
  * buzhash64 (improved), buzhash (as in borg 1.x)
  * toeplitz-aes / rabin-aes / goldilocks-aes (keyed, AES-based; they make
    chunk-boundary fingerprinting attacks much harder)
  * fixed (fixed blocksize, optionally with a differently sized header)

**Speed**
  * performance-critical code (chunking, compression, encryption) is
    implemented in C/Cython, using SIMD (NEON / AVX2 / AVX-512) where that
    is a win - the scan kernel is picked per platform
  * local caching
  * quick detection of unmodified files

**Data encryption**
    All data can be protected client-side using 256-bit authenticated encryption
    (AES-OCB or chacha20-poly1305), ensuring data confidentiality, integrity and
    authenticity.

    If you do not need confidentiality, there are also modes that only
    authenticate (``authenticated-sha256`` / ``authenticated-blake3``) and
    modes that do neither (``none-sha256`` / ``none-blake3``).

**Hashing**
    You can choose between HMAC-SHA256 and Blake3.

**Obfuscation**
    Optionally, Borg can actively obfuscate, e.g., the size of files/chunks to
    make fingerprinting attacks more difficult.

**Compression**
    All data can be optionally compressed:

    * lz4 (super fast, low compression)
    * zstd (wide range from high speed and low compression to high compression
      and lower speed)
    * zlib (medium speed and compression)
    * lzma (low speed, high compression)

**Off-site backups**
    Borg can store data on any remote host accessible via misc. protocols:

    * ``rest://`` (REST-http-over-stdio-over-ssh) and ``http(s)://``
      (REST-http-over-tcp). Significant performance gains can be achieved with
      these by having a remote agent (borg must be installed on the repo server).
    * ``sftp://``
    * ``s3://`` / ``b2://``
    * ``rclone:`` - lots of protocols / providers supported by rclone

**Backups mountable as file systems**
    Backup archives are mountable as user-space file systems for easy interactive
    backup examination and restores (e.g., by using a regular file manager).

**Easy installation on multiple platforms**
    We offer single-file binaries that do not require installing anything -
    you can just run them on these platforms:

    * Linux (x86_64, arm64)
    * macOS (arm64, x86_64)
    * FreeBSD (x86_64)
    * Windows (x86_64, MSYS2/MINGW borg.exe, experimental)

    Borg also runs on these platforms, but you need to install it from a
    package or from source:

    * NetBSD (xattrs, but no ACLs)
    * OpenBSD (no xattrs/ACLs support)
    * illumos / Solaris (xattrs, but no ACLs)
    * Cygwin (experimental)
    * Windows Subsystem for Linux (WSL) on Windows 10/11 (experimental)

**Free and Open Source Software**
  * security and functionality can be audited independently
  * licensed under the BSD (3-clause) license, see `License`_ for the
    complete license

Easy to use
~~~~~~~~~~~

For ease of use, set the BORG_REPO environment variable::

    $ export BORG_REPO=/path/to/repo

Create a new backup repository (see ``borg repo-create --help`` for encryption options)::

    $ borg repo-create --encryption=aes256-ocb --key-location=repokey

Create a new backup archive::

    $ borg create docs ~/Documents

Now do another backup, just to show off the great deduplication::

    $ borg create -v --stats docs ~/Documents
    Creating archive "docs" in repository /path/to/repo
    Repository: /path/to/repo
    Archive name: docs
    Archive fingerprint: 7714aef97c1a24539cc3dc73f79b060f14af04e2541da33d54c7ee8e81a00089
    Time (nominal): Mon, 2026-08-03 19:57:35 +0200
    Time (start):   Mon, 2026-08-03 19:57:35 +0200
    Time (end):     Mon, 2026-08-03 19:57:35 +0200
    Duration: 0.01 seconds
    Number of files: 24
    Original size: 29.73 MB
    Deduplicated size: 520 B
    ...


For demo videos, check out our homepage: https://www.borgbackup.org/#demo

Helping, donations and bounties, becoming a Patron
--------------------------------------------------

Your help is always welcome!

Spread the word, give feedback, help with documentation, testing or development.

You can also give monetary support to the project, see here for details:

https://www.borgbackup.org/#fund

Thanks to everyone who has helped so far - these are the people with the most
commits, see `FAME.md <https://github.com/borgbackup/borg/blob/master/FAME.md>`_
for the full list:

.. image:: https://raw.githubusercontent.com/borgbackup/borg/master/FAME.svg
   :alt: Borg contributors by commit count
   :target: https://github.com/borgbackup/borg/blob/master/FAME.md

Links
-----

* `Main website <https://www.borgbackup.org/>`_
* `Releases <https://github.com/borgbackup/borg/releases>`_,
  `PyPI packages <https://pypi.org/project/borgbackup/>`_ and
  `Changelog <https://github.com/borgbackup/borg/blob/master/docs/changes.rst>`_
* `GitHub <https://github.com/borgbackup/borg>`_ and
  `Issue tracker <https://github.com/borgbackup/borg/issues>`_.
* `Web chat (IRC) <https://web.libera.chat/#borgbackup>`_ and
  `Mailing list <https://mail.python.org/mailman/listinfo/borgbackup>`_
* `License <https://borgbackup.readthedocs.io/en/master/authors.html#license>`_
* `Security contact <https://borgbackup.readthedocs.io/en/master/support.html#security-contact>`_

Compatibility notes
-------------------

EXPECT THAT WE WILL BREAK COMPATIBILITY REPEATEDLY WHEN MAJOR RELEASE NUMBER
CHANGES (like when going from 0.x.y to 1.0.0 or from 1.x.y to 2.0.0).

NOT RELEASED DEVELOPMENT VERSIONS HAVE UNKNOWN COMPATIBILITY PROPERTIES.

THIS IS SOFTWARE IN DEVELOPMENT, DECIDE FOR YOURSELF WHETHER IT FITS YOUR NEEDS.

Security issues should be reported to the `Security contact`_ (or
see ``docs/support.rst`` in the source distribution).

.. start-badges

|doc| |build| |coverage| |bestpractices|

.. |doc| image:: https://readthedocs.org/projects/borgbackup/badge/?version=master
        :alt: Documentation
        :target: https://borgbackup.readthedocs.io/en/master/

.. |build| image:: https://github.com/borgbackup/borg/actions/workflows/ci.yml/badge.svg?branch=master
        :alt: Build Status (master)
        :target: https://github.com/borgbackup/borg/actions

.. |coverage| image:: https://codecov.io/github/borgbackup/borg/coverage.svg?branch=master
        :alt: Test Coverage
        :target: https://codecov.io/github/borgbackup/borg?branch=master

.. |bestpractices| image:: https://www.bestpractices.dev/projects/271/badge
        :alt: Best Practices Score
        :target: https://www.bestpractices.dev/projects/271

.. end-badges
S
Description
Deduplicating archiver with compression and authenticated encryption.
Readme BSD-3-Clause
56 MiB
Languages
Python 89.2%
Cython 6.4%
C 2.5%
HTML 1.5%
Shell 0.4%