June 2022
Yet Another Rsync is a file synchronization and backup tool
yarsync [-h] [--config-dir DIR] [--root-dir DIR] [-q | -v] command [args]
yarsync is a wrapper around rsync to store its configuration and synchronize repositories with the interface similar to git. It is efficient (files in the repository can be removed and renamed freely without additional transfers), distributed (several replicas of the repository can diverge, and in that case a manual merge is supported), safe (it takes care to prevent data loss and corruption) and simple (see this manual).
To create a new repository, enter the directory with its files and type
yarsync initThis operation is safe and will not affect existing files (including configuration files in an existing repository). Alternatively, run init inside an empty directory and add files afterwards. To complete the initialization, make a commit:
yarsync commit -m "Initial commit"commit creates a snapshot of the working directory, which is all files in the repository except yarsync configuration and data. This snapshot is very small, because it uses hard links. To check how much your directory size has changed, run du(1).
Commit name is the number of seconds since the Epoch (integer Unix time). This allows commits to be ordered in time, even for hosts in different zones. Though this works on most Unix systems and Windows, the epoch is platform dependent.
After creating a commit, files can be renamed, deleted or added. To see what was changed since the last commit, use status. To see the history of existing commits, use log.
Hard links are excellent at tracking file moves or renames and storing accidentally removed files. Their downside is that if a file gets corrupt, this will apply to all of its copies in local commits. The 3-2-1 backup rule implies that one should have at least 3 copies of data, so let us add a remote repository "my_remote":
yarsync remote add my_remote remote:/path/on/my/remoteFor local copies we still call the repositories "remote", but their paths will be local:
yarsync remote add my_drive /mnt/my_drive/my_repoThis command only updated our configuration, but did not make any changes at the remote path (which may not exist). To make a local copy of our repository, run
yarsync clone . /mnt/my_drive/my_repoclone copies all repository data (except the configuration) to the new replica and adds the original repository to its remotes with the name "origin". To copy the repository to a remote host, just copy its files (preserving hard links):
rsync -avHP ./ remote:/path/on/my/remoteHere '-H' stands for hard links. If the first path ends with a slash, data will be copied to "remote/", otherwise to its subdirectory. Similarly, one can copy a repository from a remote: just change the order of paths and don’t forget about the slash. To check that we set up the repositories correctly, make a dry run with '-n':
yarsync push -n my_remoteIf there are no errors and no file transfers, then we have a functioning remote. We can continue working locally, adding and removing files and making commits. When we want to synchronize repositories, we push the changes to or pull them from a remote (first with a --dry-run). This is the recommended workflow, and if we work on different repositories in sequence and always synchronize changes, our life will be easy. Sometimes, however, we may forget to synchronize two replicas and they will end up in a diverged state; we may actually change some files or find them corrupt. Solutions to these problems involve user decisions and are described in pull and push options.
| --help, -h | show help message and exit |
| --config-dir=DIR | path to the configuration directory |
| --root-dir=DIR | path to the root of the working directory |
| --quiet, -q | decrease verbosity |
| --verbose, -v | increase verbosity |
| checkout | restore the working directory to a commit |
| clone | clone a repository into a new directory |
| commit | commit the working directory |
| diff | print the difference between two commits |
| init | initialize a repository |
| log | print commit logs |
| pull | get data from a source |
| push | send data to a destination |
| remote | manage remote repositories |
| show | print log messages and actual changes for commit(s) |
| status | print updates since last commit |
All commands support the --help option. Commands that can change the repository also support the --dry-run option.
yarsync checkout [-h] [-n] commit
Restores the working directory to its state during commit. WARNING: this will overwrite the working directory. Make sure that all important data is committed. Make a dry run first with -n.
If not the most recent commit was checked out, the repository HEAD (in git terminology, see git-checkout(1)) becomes detached, which prevents such operations as pull or push. To advance the repository to its correct state, check out the last commit or make a new one.
yarsync clone [-h] [-o origin] [-n clone] repository directory
Clones a repository to a directory. Non-existent paths will be created. Non-empty directories will be unchanged (and an error will be issued).
Note that only data (working directory, commits and logs, not configuration files) will be cloned. This command will refuse to clone a repository with a filter (see SPECIAL REPOSITORIES).
yarsync commit [-h] [-m message]
Commits the working directory (makes its snapshot). See QUICK START for more details on commits.
yarsync diff [-h] commit [commit]
Prints the difference between two commits (from old to the new one, the order of arguments is unimportant). If the second commit is omitted, compares commit to the most recent one. See status for the output format.
yarsync init [-h] [reponame]
Initializes a yarsync repository in the current directory. Creates a configuration folder with repository files. Existing configuration and files in the working directory are unchanged. Create a first commit for the repository to become fully operational.
yarsync log [-h] [-n number] [-r]
Prints commit logs (from newest to oldest), as well as synchronization information when it is available. To see changes in the working directory, use status.
To print information about the three most recent commits, use
yarsync log -n 3yarsync pull [-h] [-f | --new | -b | --backup-dir DIR] [-n] [--overwrite] source
Gets data from a remote source. The difference between pull and push is mostly only the direction of transfer.
pull and push bring two repositories into the same state. They synchronize the working directory, that is they add to the destination new files from source, remove those missing on source and do all renames and moves of previously committed files efficiently. This is done in one run, and these changes apply also to logs and commits. In most cases, we do not want our existing logs and commits to be removed though. By default, several checks are made to prevent data loss:
- local has no uncommitted changes,
- local has not a detached HEAD,
- local is not in a merging state,
- destination has no commits missing on source.If any of these cases is in effect, no modifications will be made. Note that the remote may have uncommitted changes itself: always make a dry run with -n first!
To commit local changes to the repository, use commit. HEAD commit could be changed during checkout (see its section for the solutions). If the destination has commits missing on source, there are two options: to --force changes to the destination (removing these commits) or to merge changes inside the local repository with pull --new.
If we pull new commits from the remote, this will bring repository into a merging state. Merge will be done automatically if the last remote commit is among local ones (in that case only some older commits were transferred from there). If some recent remote commits are not present locally, however, this means that histories of the repositories diverged, and we will need to merge them manually. After we have all local and remote commits and the union of the working directories in our local repository, we can safely choose the easiest way for us to merge them. To see the changes, use status and log. For example, if we added a file in a remote_commit before and it was added now, we can just commit the changes. If we have made many local changes, renames and removals since then, we may better checkout our latest commit (remember that all files from the working directory are present in commits, so it is always safe) and link the new file to the working directory:
ln .ys/commits/<remote_commit>/path/to/file .(it can be moved to its subdirectory without the risk of breaking hard links). If the remote commit was actually large, and local changes were recent but small, then we shall check out the remote commit and apply local changes by hand. After our working directory is in the desired state, we commit changes and the merge is finished. The result shall be pushed to the remote without problems.
Do not remove local data that is missing on source. While this option can return deleted or moved files back to the working directory, it also adds remote logs and commits that were missing here (for example, old or unsynchronized commits). A forced push to the remote could remove these logs and commits, and this option allows one to pull them first to the local repository.
After pull --new the local repository can enter a merging state. See pull description for more details.
Changed local files are put into a directory DIR preserving their relative paths. DIR can be an absolute path or relative to the root of the repository. In contrast to --backup, --backup-dir does not change resulting file names.
This option is convenient for large file trees, because it recreates the existing file structure of the repository (one doesn’t have to search for new backup files in all subdirectories). For current rsync version, the command
yarsync pull --backup-dir BACKUP <remote>will copy updated files from the remote and put them into the directory "BACKUP/BACKUP" (this is how rsync works). To reduce confusion, make standard pull first (so that during the backup there are only file updates).
This option is available only for pull, because it is assumed that the user will apply local file changes after backup. For example, suppose that after a pull --backup one gets files a and a~ in the working directory. One should first see, which version is correct. If it is the local file a~, then the backup can be removed:
mv a~ a*By local we mean the one hard linked with local commits (run ls -i to be sure). If the remote version is correct though, you need first to overwrite the local version not breaking the hard links. This can be done with an rsync option "--inplace":
rsync --inplace a a~
mv a~ a
# check file contents and the links
ls -i a .ys/commits/*/aFor a --backup-dir and for longer paths these commands will be longer. Finally, if you need several versions, just save one of the files under a different name in the repository.
After you have fixed all corrupt files, push them back to the remote. See the --overwrite option for propagation of file changes.
yarsync push [-h] [-f] [-n] [--overwrite] destination
Sends data to a remote destination. See pull for more details and common options.
yarsync remote [-h] [-v] [command]
Manages remote repositories configuration. By default, prints existing remotes. For more options, see .ys/config.ini in the FILES section.
yarsync remote add [-h] repository path
Adds a new remote. repository is the name of the remote in local yarsync configuration (as it will be used later during pull or push). path has a standard form [user@]host:[path] for an actually remote host or it can be a local path. Since yarsync commands can be called from any subdirectory, local path should be absolute. Tilde for user’s home directory '~' in paths is allowed.
yarsync remote rm [-h] repository
Removes an existing repository from local configuration.
Prints remote repositories. Default.
yarsync show [-h] commit [commit …]
Prints log messages and actual changes for commit(s). Changes are shown compared to the commit before commit. For the output format, see status. Information for several commits can be requested as well.
yarsync status [-h]
Prints working directory updates since the last commit and the repository status. If there were no errors, this command always returns success (irrespective of uncommitted changes).
The output for the updates is a list of changes, including attribute changes, and is based on the format of rsync --itemize-changes. For example, a line
.d..t...... programming/means that the modification time 't' of the directory 'd' programming/ in the root of the repository has changed (files were added or removed from that). All its other attributes are unchanged ('.').
The output is an 11-letter string of the format "YXcstpoguax", where 'Y' is the update type, 'X' is the file type, and the other letters represent attributes that are printed if they were changed. For a newly created file these would be '+', like
>f+++++++++ /path/to/fileThe attribute letters are: checksum, size, modification time, permissions, owner and group. u can be in fact use (access) or creation time, or both. a stands for ACL, and x for extended attributes. Complete details on the output format can be found in the rsync(1) manual.
A detached repository is one with the yarsync configuration directory outside the working directory. To use such repository, one must provide yarsync options --config-dir and --root-dir with every command (alias(1p) may be of help). To create a detached repository, use init with these options or move the existing configuration directory manually. For example, if one wants to have several versions of static Web pages, they may create a detached repository and publish the working directory without the Web server having access to the configuration. Alternatively, if one really wants to have both a continuous synchronization and yarsync backups, they can move its configuration outside, if that will work. Commits in such repositories can be created or checked out, but pull or push are currently not supported (one will have to synchronize them manually). A detached repository is similar to a bare repository in git, but usually has a working directory.
A repository with a filter can exclude (disable tracking) some files or directories from the working directory. This may be convenient, but makes synchronization less reliable, and such repository can not be used as a remote. See rsync-filter in the FILES section for more details.
All yarsync repository configuration and data is stored in the hidden directory .ys under the root of the working directory. If the user no longer wants to use yarsync and the working directory is in the desired state, they can safely remove the .ys directory.
Note that only commits and logs (apart from the working directory) are synchronized between the repositories. Each repository has its own configuration and name.
Contains names and paths of remote repositories. This file can be edited directly or with remote commands according to user’s preference.
yarsync supports synchronization with only existing remotes. A simple configuration for a remote "my_remote" could be:
[my_remote]
path = remote:/path/on/my/remoteSeveral sections can be added for more remotes. An example (non-effective) configuration is created during init. Note that comments in config.ini can be erased during remote {add,rm}.
Since removable media or remote hosts can change their paths or IP addresses, one may use variable substitution in paths:
[my_drive]
path = $MY_DRIVE/my_repoFor the substitutions to take the effect, export these variables before run:
$ export MY_DRIVE=/run/media/my_drive
$ yarsync push -n my_driveIf we made a mistake in the variable or path, it will be shown in the printed command. Always use --dry-run first to ensure proper synchronization.
Another yarsync remote configuration option is host. If both path and host are present, the effective path will be their concatenation "<host>:<path>". Empty host means local host and does not prepend the path.
It is possible to set default host for each section from the section name. For that, add a default section with an option host_from_section_name:
[DEFAULT]
host_from_section_nameEmpty lines and lines starting with '#' are ignored. Section names are case-sensitive. White spaces in a section name will be considered parts of its name. Spaces around '=' are allowed. Full syntax specification can be found at https://docs.python.org/3/library/configparser.html.
Contains the repository name, which is used in logs and usually coincides with the remote name (how local repository is called on remotes). The name can be set during init or edited later.
It is recommended (but not required) to have different names for the repository replicas on different hosts or devices. For example, if one has repositories "programming/" and "music/" on a laptop "my_host", their names would probably be "my_host", and the names of their copies on an external drive could be "my_drive" (this is different from git, which uses only the author’s name in logs). If one never creates commits directly on "my_drive", these names can be empty.
If the repository name is missing (empty or no file), host name will be used. If there is an error getting the host name during commit, provide the name in the repository.txt.
Contains rsync filter rules, which effectively define what data belongs to the repository. The rsync-filter does not exist by default, but can be added for flexibility.
For example, the author has a repository "~/work", but wants to keep his presentations in "tex/" in a separate repository. Instead of having a different directory "~/work_tex", he adds such rules to rsync-filter:
# all are in git repositories
- /repos
# take care to sync separately
- /texIn this way, "~/work/tex" and contained git repositories will be excluded from "~/work" synchronization. Lines starting with '#' are ignored, as well as empty lines. To complicate things, one can include a subdirectory of "tex" into "work" with an include filter '+'. For complete details, see FILTER RULES section of rsync(1).
While convenient for everyday use, filters make backup more difficult. To synchronize repository with them, one has to remember that it has subdirectories that need to be synchronized too. If the remote repository had its own filters, that would make synchronization even more unreliable. Therefore filters are generally discouraged: pull and push ignore remote filters (make sure you synchronize only from a repository with filters), while clone refuses to copy a repository with rsync-filter.
If the command could be run successfully, a zero code is returned. Invalid option code is returned for mistakes in command line argument syntax. Configuration error can occur when we are outside an existing repository or a yarsync configuration file is missing. If the repository is correct, but the command is not allowed in its current state (for example, one can not push or pull when there are uncommitted changes or add a remote with an already present name), the command error is returned. It is also possible that a general system error, such as a keyboard interrupt, is raised in the Python interpreter.
In case of rsync errors, its error code is returned.
To check that your clocks (used for properly ordering commits) at different hosts are synchronized well enough, run
python -c 'import time; print(time.time())'To make sure that the local repository supports hard links instead of creating file copies, test it with
du -sh .
du -sh .ys(can be run during pull or clone if they take too long). The results must be almost the same. If not, you may not use yarsync on this file system, have large deleted files stored in old commits or you may have subdirectories excluded with a filter (see SPECIAL REPOSITORIES section).
To test that a particular file "a" was hard linked to its committed versions, run
ls -i a .ys/commits/*/aIf all is correct, their inodes must be the same.
Hard links may be broken in a cloned git repository (as it happens with yarsync tests), because git does not preserve them. To fix hard links for the whole repository, run hardlink(1) in its root.
rsync(1)
The yarsync page is https://github.com/ynikitenko/yarsync.
Requires a filesystem with hard links, rsync version at least 3.1.0 (released 28 September 2013) and Python >= 3.6.
Always do a --dry-run before actual changes.
Occasionally Python errors are raised instead of correct return codes. Please report any bugs or make feature requests to https://github.com/ynikitenko/yarsync/issues.
Copyright © 2021-2022 Yaroslav Nikitenko. License GPLv3: GNU GPL
version 3 https://gnu.org/licenses/gpl.html.
This is free software: you are free to change and redistribute it. There
is NO WARRANTY, to the extent permitted by law.