File Replicator

Replicate files one-way to another computer e.g. for remote development.

A key use-case is to keep in sync a directory of development files from a computer on which the files are edited with a copy of those files in a docker container running on a remote docker host.

Tested and known to work between two Linux machines. Support for developing on macOS coming...

Installation

Dependencies are:

• Python 3 and some Python packages on the development machine.
• Ability to run a shell (bash or bash-like) on the remote machine with connected stdin.
• The tar utility (the full version, not the busybox version) on both machines.

Note that nothing is installed remotely, there are no ports to open, and the remote user only needs the ability to create the files and directories at the specified location.

So to install file-replicator on the machine with the source files to replicate:

pip install file-replicator

Nothing needs to be installed on the destination machine so long as it has bash (busybox bash is fine) and tar (gnu). Note that on alpine linux, the busybox tar is insufficient, so install gnu tar with:

apk install tar

How it works

The approach involves running a small bash program on the remote (destination) end which is able to add/update new files in (potentially) new directories. It receives these files over stdin using the tar format (binary).

The controlling (source) end then simply sends files over to the stdin of the receiving bash program, which pipes them through tar to unpack them again. Note that gnu tar is able to extract from non-blocking file descriptor (as well as blocking), which means it keeps trying until it has all the data. NB the busybox tar does not have this behaviour.

Establishing the connection to the remote end is outside the remit of the tool, but file-replicator requires as an argument the command to make such a connection. See examples below.

Once a connection has been made, two phases of operation occur:

1. first, recursively walk a source tree of files and sending all of them over the wire to the destination
2. then, watch for changes or new files and directories before sending them over the wire to the destination

So there is no "difference algorithm" like rsync, no attempt to compress (although of course the connection could already be compressing e.g. if over ssh), the connection is made entirely using standard means like ssh and docker, no ports to open, and even the bash program on the remote end is sent over every time so nothing is installed remotely.

This is sufficient for editing code on a local computer and automatically replicating them to a remote server or docker container whenever a file is created or modified.

Usage and examples

See help with file-replicate --help:

Usage: file-replicator [OPTIONS] SRC_DIR DEST_PARENT_DIR
                       [CONNECTION_COMMAND]...

  Replicate files to another computer e.g. for remote development.

  SRC_DIR is the source directory on this machine.

  DEST_PARENT_DIR is the (absolute) destination parent directory on the
  remote machine accessed using the CONNECTION_COMMAND.

  The CONNECTION_COMMAND must result in a running instance of bash ready to
  receive commands on stdin.

  Example CONNECTION_COMMANDS include:

      ssh some.host.com bash

      docker exec -i my_container bash

      docker-compose exec -T my_container bash

  So a full use of the tool might look like:

      file-replicator my_code_dir /home/code -- docker exec -i a_container
      bash

  (the use of "--" prevents any further processing of command line arguments
  by file-replicator, leaving them all for docker)

  Initially, all files and required directories are recursively copied. Then
  it waits for changes before copying each modified or new file. This can be
  modified with the switches.

  Note that empty directories are not replicated until they contain a file.

  Lastly, the only time the tool deletes files or directories is if called
  with the optional --clean-out-first switch.

Options:
  --clean-out-first               Optionally start by cleaning out the
                                  destination directory.
  --with-initial-replication / --no-initial-replication
                                  Perform (or not) an initial replication of
                                  all files.
  --replicate-on-change / --no-replicate-on-change
                                  Perform (or not) a wait-for-change-and-
                                  replicate cycle.
  --gitignore / --no-gitignore    Use .gitignore (or not) to filter files.
  --version                       Show the version and exit.
  --help                          Show this message and exit.

For example, to replicate files from local directory my_project_dir to directory /home/code/my_project_dir on remote machine called my.server.com:

file-replicator my_project_dir /home/code ssh my.server.com bash

As another example, to replicate files from local directory my_project_dir to directory /home/code/my_project_dir in a running docker container called my_container on a potentially remote host (depending upon the DOCKER* environment variables e.g. as set by docker-machine eval):

file-replicator my_project_dir /home/code -- docker exec -i my_container bash

Or to do the same but using docker-compose instead:

file-replicator my_project_dir /home/code -- docker-compose exec -T my_container bash

Lastly, as a degenerate example which doesn't actually connect to a remote machine at all but replicates into the local /tmp/my_project_dir:

file-replicator my_project_dir /tmp bash

The unit tests use this degenerate approach to test the tool.

Limitations

Due to limitations with inotify (race conditions around watching for changes in newly created directories), it is possible that the watching-for-changes phase becomes out of step. In which case, just restart the whole program. The tool includes some self-restarting behaviour, but ultimately a full restart may sometimes be needed.

Information printed to stdout indicates when this happens.

Tests

============================= test session starts ==============================
platform linux -- Python 3.6.7, pytest-3.10.1, py-1.7.0, pluggy-0.8.0 -- /home/tcorbettclark/.cache/pypoetry/virtualenvs/file-replicator-py3.6/bin/python
cachedir: .pytest_cache
rootdir: /home/tcorbettclark/code/file-replicator, inifile:
collecting ... collected 8 items
tests/test_lib.py::test_empty_directories_are_copied PASSED                [ 12%]
tests/test_lib.py::test_copy_one_file PASSED                               [ 25%]
tests/test_lib.py::test_copy_file_with_unusual_characters_in_name PASSED   [ 37%]
tests/test_lib.py::test_make_missing_parent_directories PASSED             [ 50%]
tests/test_lib.py::test_replicate_all_files PASSED                         [ 62%]
tests/test_lib.py::test_detect_and_copy_new_file PASSED                    [ 75%]
tests/test_lib.py::test_detect_and_copy_modified_file PASSED               [ 87%]
tests/test_lib.py::test_detect_and_copy_new_file_in_new_directories PASSED [100%]
=========================== 8 passed in 3.98 seconds ===========================

Contributions

Pull-requests are welcome! Please consider including tests and updating docs at the same time.

The package is maintained using poetry (https://poetry.eustace.io) and pyenv (https://github.com/pyenv/pyenv).

The code is formatted using black (https://black.readthedocs.io/en/stable) and isort (https://github.com/timothycrosley/isort).

It is tested using pytest (https://pytest.org).

Commit checklist

1. check version both in pyproject.toml and file_replicator/__init__.py
2. git tag
3. isort -rc .
4. black .
5. pytest -v
6. update this README.md with the latest output from the tests
7. update this README.md with the latest output from the --help option