If you just want the How To, skip the Why and Rsync sections and jump directly to the Steps section.

Why?

I am a bioinformatician constantly working on a server. For my current project (master and thesis), I wrote plenty of scripts in a directory named —guess what— scripts/.

But what happens when the server is down? (And that happens more often that you’d wish, well you know what I mean)

I love git, and had “version-controled” my ~/scripts/ directory for a few months, but without setting a remote. The reason is that our system administrator encourages us not to send our work on sites like github, and I couldn’t even negociate for a (free) private repository on bitbucket.

An attempt with rsync

It’s gonna be easier if you know the names of the two machines we are talking about so let me introduce you to these charming beasts:

• mydawg is the computer on my desk, the one that is less likely to crash;
• amphisbaena is the server (hidden reptilian entity living underground).

Also you’ll need my username on both machine, let’s say me.

Ok, so I first thought I could simply use rsync and cron to periodically synchronize between my working machine (mydawg) and the server (amphisbaena).

I was ambitious and wanted to be able to work both from mydawg and amphisbaena: therefore I need to avoid overwriting more recent changes on any side.

I made this script on mydawg in the directory scripts/:

#!/usr/bin/bash
# Name of this script: sync.sh

# NOTE: the '/' at the end is important. see man rsync.
local="./"
remote="amphisbaena:programTestingArea/synced_with_mydawg/"

# download
rsync -ravz --update --exclude "sync.sh" --exclude ".sync.sh.swp" $remote $local
# upload
rsync -ravz --update --exclude "sync.sh" --exclude ".sync.sh.swp" $local $remote

Explanation:

--update is to “skip files that are newer on the recever”;

-a is an alias for many useful option that you wanna use when doing backups;

-r copy recursively in subdirectories. For some obscure reason, even though -a is an alias for -rlptgoD, it seemed to me that it doesn’t work as expected if you don’t add -r…

-v for verbose;

-z to compress during transfer.

That works, however, what happens when you delete or move a file on one side? It’ll come back at the next syncing, so you’ll actually need to execute the delete or move on both sides. You can’t counter this by using the option --delete because that would delete newly created files on one side and that’s not what you want… Or if you use --delete only in the “download” command, you can’t create new files on mydawg. If you happen to forget that, your new creation will be soon deleted.

Plus, I will slip to our transition by saying that you must not forget to sync before working or you could end up with conflicts in case you made new stuff on both sides.

All that verbiage lead us to consider git as the ideal solution.

Git: steps

log on the server:

ssh amphisbaena

Create the repository directory:

mkdir -P ~/mygit/scripts_amphisbaena.git

init the repo as a sharing repo (not working). You basically can’t have a working tree in a repo designed for sharing, that’s explained here. Took me some time to confirm that intuition. That’s why here you use the --bare option:

git init --bare

Now set up this repository as the remote for your existing working repository:

cd ~/scripts/
git remote add origin file:///users/me/mygit/scripts_amphisbaena.git
git push -u origin master

Side note: you need to use -u (--set-upstream) with git push because it will later allow you to simply type git pull instead of git pull origin master. If you forget it the first time, you can still do your git push -u later.

Al’right, now your sharing repository is set up on the server. You can go on your local machine and clone it using the ssh protocole:

# This is performed on `mydawg`
git clone ssh://me@amphisbaena/users/me/mygit/scripts_amphisbaena.git

And that was all! If you searched the net for “how to set up a git server”, you would have found commands to create a “git” user and so on, actions that require sudo privileges so you usually can’t perform that on a team server. It may be obvious to you that the above simple procedure was sufficient, but wasn’t completely clear to me when researching the subject.