mighdoll / jsync

Welcome to jSync - transparent object replication over json

This is a technology preview release -- you can see the api, peek at the code,
and try some small test cases.

Here's how to try it:
$ mvn install
$ cd server && mvn jetty:run
$ firefox http://localhost:8080/jsync/test/tests.html

Reading about jSync
* See the doc directory for some early docs
* The server code is in: server/src/main/scala/com/digiting/sync
* The javascript code is in: server/src/main/webapp/jsync
* Poke around, there are more comments in the code.

----------------

The goal is to enable easy-to-use object replication between
loosely connected machines.  A browser and a server can share a pool
of objects.  If the browser program makes a local change to any object
in the pool, the change will propagate to the server.  And vice versa.

What this enables:
* Way simple persistence for javascript.  
    Change a few lines and your javascript objects can flow to the server
    and from there to MySQL or whatever.

* Rich models
    It's easy to communicate lots of different kinds of data and
    relationships, which is especially handy for complex models and
    rich clients.

* Observation api.  
    Use the observation api to do MVC/data binding style programming.  And 
    then support for network delivered changes comes for free.

A few key features: 
* Transactions 
    Related changes to objects are normally bundled into atomic groups.  
    Changes are applied in bulk, so that programs don't see inconsistent state.

* Syncable objects are normal objects... (ideally.)
    In practice, a few compromises are required to make sync work in
    javascript and scala.  The current javascript api requires a special
    constructor syntax, and using setter functions (on old browsers).
    The scala api is TBD, but may require goat herding skills.

* Collections 
    Starting with three mutable collections:  Set, Map, and LinkedList. 
    
* Asynchronous 
    Neither the browser the server can afford to wait on the other.

* Observation api 
    Watch() any syncable object or collection for changes.
    
* Subscription api 
    Pools of shared objects are identified by subscription root object.
    The subscription contains every object referenced directly or
    indirectly by the root object.

* JSON protocol 
    The current protocol sketch is simple and uses JSON, so it's trendy.
    It's reasonably efficient too, in that only the changes to objects and
    collections are sent.

* Conflict notification, eventual consistency
    Occasionally, two programs will change the same thing before
    synchronization catches up.  It's important to detect the conflict
    case to ensure that the two copies converge.  And although Thomas's
    last one wins is usually fine, it's worth giving the program a chance
    to try some fancier recovery.
    

Future:
* Recovery 
    Using other replicated stores to identify loss of sync and potentially
    to bootstrap recovery.   e.g. if the browser crashes, use the server
    to refresh.  If a server fails, detect whether the browser is in
    sync with the replacement server.

* Server to server
    The design and protocol is pretty symmetric anyway, so it shouldn't
    be too much work to support a server as a client of another server.

* Pageable collections
    Support for large collections on the server that are viewable by the
    client section by section, but not copied in their entirety.