git-annex rclone special remote

Users are urged to upgrade to version 0.4 or higher immediately. Some ways of using earlier versions could result in data loss. more information here

This wrapper around rclone makes any destination supported by rclone usable with git-annex.

Cloud storage providers supported by rclone currently include:

Google Drive
Amazon S3
Openstack Swift / Rackspace cloud files / Memset Memstore
Dropbox
Google Cloud Storage
~~Amazon Cloud Drive~~ (issue #22)
Microsoft One Drive
Hubic
Backblaze B2
Yandex Disk

Installation

Install git-annex
Install rclone into your $PATH, e.g. /usr/local/bin
Copy git-annex-remote-rclone into your $PATH

Requirements

Because rclone sometimes changes its output, updates to this software may be required for compatibility. Version 1.45 of rclone introduces a more stable JSON output mode on some commands and for that reason is the minimum supported version.

A periodic continuous integration process tests compatibility against the latest stable release of git-annex and the following versions of rclone:

the latest stable release at time of testing
1.59.2 (the latest stable release at time of writing)
1.58.1
1.53.3
1.45 (Linux only)

The build badge above is linked to this CI process.

Usage

Configure rclone remote: rclone config.
If your rclone configuration is password-protected then export the RCLONE_CONFIG_PASS environment variable in the shell where you will execute git annex command:

export RCLONE_CONFIG_PASS=your_password_here

or prefix `git annex` command with the environment variable:

RCLONE_CONFIG_PASS=your_password_here git annex ...

Pull requests proposing a more secure or easier to use approach to password-protected rclone configurations are welcome.

Create a git-annex repository (walkthrough)
Choose a repository layout. If you are having difficulty choosing, the lower layout is recommended. Supported layouts:
- lower - A two-level lower case directory hierarchy is used (using git-annex's DIRHASH-LOWER MD5-based format). This choice requires git-annex 6.20160511 or later.
- directory - A two-level lower case directory hierarchy is used, along with the key name as a 3rd level nested directory. This choice requires git-annex 6.20160511 or later.
  - Some cloud providers require traversing a tree to request a file. The additional nested directory may cause a small performance loss for remote operations.
  - Known compatible remotes: directory, rsync
- nodir - No directory hierarchy is used.
  - On systems which are designed to efficiently deal with many objects in a single "directory" or "path", this is the simplest and most efficient layout.
  - Known compatible remotes: Thomas Jost's Hubic remote when a swift container other than default is used.
- mixed - A two-level mixed case directory hierarchy is used (using git-annex's DIRHASH format).
  - This layout may cause problems when used on filesystems and cloud storage providers that are case-insensitive.
  - Known compatible remotes: Thomas Jost's Hubic remote when the default swift container is chosen.
- frankencase - A two-level lower case directory hierarchy is used (using git-annex's DIRHASH format, with all characters translated to lower case)
  - This layout should not be used except if you already have a legacy remote using this layout and do not wish to migrate.
  - This was the only available layout in early versions of this remote, up to release v0.1.
Add a remote for the provider. This example:
- Adds a git-annex remote called myacdremote
- Stores your files in an rclone remote configured with the name acd
- Uses a lower repository layout
- Stores your files in a folder/prefix called git-annex:

git annex initremote myacdremote type=external externaltype=rclone target=acd prefix=git-annex chunk=50MiB encryption=shared mac=HMACSHA512 rclone_layout=lower

The initremote command calls out to GPG and can hang if a machine has insufficient entropy. To debug issues, use the --debug flag, i.e. git-annex initremote --debug.

Choosing a Chunk Size

Choose your chunk size based on your needs. By using a chunk size below the maximum file size supported by your cloud storage provider for uploads and downloads, you won't need to worry about running into issues with file size. Smaller chunk sizes: leak less information about the size of file size of files in your repository, require less ram, and require less data to be re-transmitted when network connectivity is interrupted. Larger chunks require less round trips to and from your cloud provider and may be faster. Additional discussion about chunk size can be found here and here

Upgrading From 0.1 and Earlier

git-annex-remote-rclone now requires specifying a rclone_layout= setting. Earlier versions used the frankencase layout, which is no longer recommended.

A migration script is included which will convert your data from the frankencase layout to the lower layout.

Avoid using the remote during migration. The migration script is idempotent. It is safe to re-run it if migration is interrupted or if some move operations do not complete successfully. When running the script no longer generates any 'move' commands, all objects have been migrated to the new layout.

Implementation Note

At this time, this remote does NOT store your credentials in git-annex. Users are responsible for ensuring a ~/.rclone.conf file with valid credentials is available.

Warning

Not all of the supported cloud storage providers have been tested. While in theory any provider that works with rclone should also work with this remote, it's possible there are unforeseen integration issues - issues which might even result in data loss.

Issues, Contributing

If you run into any problems, please check for issues on GitHub. Please submit a pull request or create a new issue for problems or potential improvements.

internetimagery/git-annex-remote-rclone