spencerx / hesokuri Goto Github PK

#Hesokuri

Distributed Git repo backup and duplication daemon.

##Intro Hesokuri synchronizes one or more Git source code repositories between multiple machines on a network. It is useful in the following situations:

1. You want the peace of mind of controlling the machines where your source code is stored.
2. You control/own multiple machines and work on the same project on them.
3. You want changes on one machine to appear on all other machines as soon as they are connected.
4. There is enough disk capacity on the machines such that each repository can fit on two or more peers.
5. The data to store is confidential or sensitive so it cannot be stored on a third-party’s machine or on a traditional cloud storage system.
6. You want to use existing hardware rather than buy a dedicated backup device.

###What Hesokuri does Simply stated, as soon as you commit a change in a repository, Hesokuri attempts to push the changes to every peer that holds an instance of the repository in the configuration file.

Hesokuri pushes commits on local branches to a remote branch called X_hesokr_Y, where X is the local name of the branch and Y is the identity of the host pushing the change.

####Live-edit branches Sometimes, you will manually merge changes from X_hesokr_Y into the local branch X when you need to access them. However, if X is a live-edit branch, and it is a fast-forward of the local X branch, and the local X branch is either not checked out or it is checked out and the working area and index are clean, X will automatically be reset to the commit pointed to by X_hesokr_Y. This means if X is checked out and there are no uncommitted changes to it, your working tree will update automatically when another peer commits changes to it. This process is referred to as advancing in the source code.

You can specify which branches are live-edit in the configuration file (see the Configure section below).

####Sharing changes from other peers For every branch named X_hesokr_Y, Hesokuri will try to push it to other peers (besides Y) using the same branch name. This is useful for sharing changes between two peers that are not running at the same time, but are running at the same time as some third peer. In this case, the third peer acts as a carrier for changes that did not originate from it.

##Getting started

###Requirements All machines should have the following. Use earlier versions of each component at your own risk.

1. Hesokuri source
2. Leiningen 2.1.3 or higher
3. Java Development Kit 1.7 or higher
4. git version 1.7.10.1 or higher
5. A Unix-like OS is recommended. Hesokuri is tested on Mac OS X and Linux, but Windows is worth a try if you are feeling adventurous.
6. A static hostname or IP address. i.e. you should be able to do ping FOO from any machine and get a response, where FOO always refers to the same machine. This is string is called the identity.
7. Public-key ssh login enabled for each machine. Super-condensed directions which will work for most cases (CENTRAL is the identity of some arbitrary "central peer"):
   • Enable ssh login on each machine if it is not already
   • For each peer, run ssh-keygen -t rsa (enter an empty string when asked for a passphrase)
   • For each peer FOO (including CENTRAL), run:
     scp ~/.ssh/id_rsa.pub CENTRAL:/tmp/id_rsa.pub.FOO
   • After the previous command has been run on all peers, run the following on CENTRAL for each peer FOO:
     cat /tmp/id_rsa.pub.* > /tmp/heso_keys
   • For each peer FOO (including CENTRAL), run
     scp CENTRAL:/tmp/heso_keys /tmp/heso_keys
cat /tmp/heso_keys >> ~/.ssh/authorized_keys

###Configure On each peer, create a file at ~/.hesocfg which specifies all Git repos and peers to sync with. Each copy of the file can be the same. The file should contain a Clojure expression that is a vector of maps. Each element in the vector is a different Git repository to sync (called a source). The map is a dictionary which contains various parameters that control how the source is synchronized. It requires at least a :host-to-path entry, which must be set to a map of peer identities to local paths on each peer for the source. A source need not appear on every peer. Here is an example which demonstrates the syntax and optional source parameters:

{:comment
 ["You can put any data here as notes to maintainers of this file."]

 :sources
 [{:host-to-path {"host-1" "/home/fbar/repo1"
                  "host-2" "/home/fbar/repo1"}}

  {:host-to-path {"host-1" "/home/fbar/repo2"
                  "host-3" "/home/fbar/repo2"}
   :unwanted-branches #{"baz" "42"}}

  {:host-to-path {"host-1" "/home/fbar/repo3"
                  "host-3" "/home/fbar/repo3"}
   :live-edit-branches {:only #{"master"}}}]

The syntax is identical to Clojure literal syntax. The important constructs are:

• {} indicates a map, which contains keys and values in alternating order. Commas can be used in a map to separate key/value pairs, but they are optional.
• :foo indicates a keyword named foo, which is used for keys in maps
• #{} indicates a set
• [] indicates a vector

Note that the distinction between maps, sets, and vectors are important when writing a configuration file. The top level data structure is a map with a :sources key (required) and :comment key (optional). The value of :sources must be a vector containing source definitions, each definition being a map.

The meaning of each entry in a source definition is as follows:

• :host-to-path is a map of peer identities to the local path of that repo on the peer. Generally, you will want to make each peer have the same or similar path for a given repo.
• :unwanted-branches indicates the names of branches that should be deleted on this host. In the example, baz is an unwanted branch, so any branch named baz or baz_hesokr_* will be deleted on the host with this configuration. Such branches are deleted with git branch -D, which means that unmerged changes will be lost. Be careful using this option!
• :live-edit-branches indicates which branches are considered live-edit, which by default is only hesokuri. Live-edit branches are branches that, when a peer pushes changes to a local repo for a source, those peer's changes are merged in automatically if it is a fast-forward. If you specify {:only FOO} for this parameter, only the branches in the set FOO will be live-edit. If you specify {:except BAR}, then every branch will be live-edit except for the ones in the set BAR.

You can edit a configuration file while Hesokuri is running, and it will automatically restart with the new configuration.

An annotated, real-world configuration file may look something like this:

{:comment
 ["My machines:"
  "192.168.0.2 - the Linux laptop"
  "192.168.0.3 - the Mac desktop"
  "192.168.0.4 - the home server"]

 :sources
 [{:comment "Whiz bang project"
   :host-to-path {"192.168.0.2" "/home/johndoe/whizbang"
                  "192.168.0.3" "/Users/johndoe/whizbang"
                  "192.168.0.4" "/home/johndoe/whizbang"}
   :unwanted-branches #{"abandoned-feature1"}
   :live-edit-branches {:only #{"master" "vnext"}}}

  {:comment "Code that may come in handy later"
   :host-to-path {"192.168.0.3" "/Users/johndoe/hacks/lisp-snippits"
                  "192.168.0.4" "/home/johndoe/hacks/lisp-snippits"}
   :live-edit-branches {:except #{"private"}}}

  {:host-to-path {"192.168.0.3" "/Users/johndoe/hacks/game-engine"
                  "192.168.0.4" "/home/johndoe/hacks/game-engine"}}]}

Note that you can save the configuration file in a location other than ~/.hesocfg and set the environment variable HESOCFG to its location. This way, you can store the HESOCFG in a subdirectory and put it in a Git repo to sync (along with other miscellaneous configuration files and utilities that are shared between all systems).

If a repo or any containing directory does not exist on a peer and you start Hesokuri on it, the containing directory and repo will be created automatically. This means that, in the above example, if the hacks folder containing the two repos does not exist on 192.168.0.4, you can run Hesokuri anyway, 192.168.0.4 will create the folders and initialize empty Git repos, and other peers will push the two repos to it as soon as they establish a connection.

###Run To run, switch to the directory containing the Hesokuri source and enter lein run at the command line.

###Web interface You can go to http://localhost:8080 in a web browser on any peer to see the status of all sources and last-pushed hashes for each branch and peer. The port can be changed from the default of 8080 by setting the environment variable HESOPORT to the desired port number.

##More information

• The wiki on Github documents things that don't quite fit in the README, like the contribution process.
• Join the Google group to get tips, discuss possible improvements to the tool, etc.

##FAQ

###Where does the name of the project come from? From the Japanese word meaning "secret cash hoard." It was chosen because this tool enables a kind of "hoarding" of data on a personal machine. The name also contrasts this practice with the alternative of storing your data on a third party server, while the alternative to a hesokuri is storing your money in the family bank account.

###Does Hesokuri support synchronizing bare repositories? Hesokuri is mostly tested with non-bare repositories with their own working trees. Bare repositories should also work, assuming they are initialized manually with git init --bare in the directory specified by the configuration file.

###What happens when some peer is unreachable? Before pushing a branch to a remote peer, Hesokuri makes sure it is responsive with InetAddress.isReachable (essentially a ping). If it is not responsive, Hesokuri will not push anything to that peer. Every three minutes, Hesokuri attempts a ping and push for every peer that does not have the most recent version of a branch. You can also use the web interface to force an immediate push for peers that failed their last ping.