Re: [jgit-dev] Using JGit in Google AppEngine

[Shawn Pearce <spearce@xxxxxxxxxxx>]

On Mon, Feb 7, 2011 at 11:24, Shawn Pearce <spearce@xxxxxxxxxxx> wrote:
> On Mon, Feb 7, 2011 at 11:05, Jason Kridner <jkridner@xxxxxxxxxxxxxxx> wrote:
>> I saw some interest in this some months ago and wondered if there was
>> any resolved solution.  Does anyone have something working today?
>
> Working?  No.  But I have been working on porting JGit to use an
> Apache Cassandra cluster as its storage backend.  Most of the work is
> still a work-in-progress, you can find my generic code at
> http://egit.eclipse.org/r/2295.  Since AppEngine supports a NoSQL
> database, it may be feasible to map this work onto AppEngine's
> datastore by implementing the org.eclipse.jgit.storage.dht.spi
> interfaces.  You may also be able to take advantage of the memcache
> server in AppEngine by implementing the CacheService interface and
> wrapping the Database with a CacheDatabase.
>
> To be honest, this probably will only work for small repositories.
> Last time I looked at Google AppEngine, there are some serious
> limitations on the amount of data you can pull into the application at
> once, or how much data you can read from, or return to the client.

I probably should point out that I think org.eclipse.jgit.storage.dht
is *the* way to put Git onto a non-filesystem storage, which includes
Google AppEngine, Amazon S3, Apache Hadoop, Apache Cassandra, Redis,
etc.  And I would love to see a Google AppEngine SPI contributed back
to JGit as part of the base JGit library.  There are many compelling
uses for small Git repositories on Google AppEngine, for example the
GitHub Gist service (many small Git repositories, each their own
pastebin).  But I really doubt Google AppEngine will support the large
data sets involved for the linux-2.6 kernel repository, or the entire
Eclipse.org project space.

-- 
Shawn.