Repository locking with clustering

[Jon Walker <jw...@shokker.net>]

I have been playing around with Jackrabbit 1.3 trying to get a feeling for 
the functionality as research for a project I'm working on. We are looking 
to wrap the Jackrabbit functionality into webservices so we can provide 
the repository functionality to multiple applications on multiple 
servers, without using the (admittedly slow) RMI.  We 
also want the webservices to be on redundant servers to prevent a single 
point of failure.

I've been looking at the clustering feature that was added in, but am 
having trouble with the locking.  I have 2 Model 1 deployments set up 
using clustering (each with a unique ID).  Through shared drives, they are 
accessing the same repository location (unique configurations for the 
clustering) and I have SimpleDBPersistence to SQL Server for persistence.

On first access to each instance, the webservices attempt to start the 
repository.  The first server to be hit will create a Repository instance 
and create the .lock file in the repository's home directory.  When the 
other server gets hit and attempts to start the repository, it throws an 
exception saying "The repository home at [path] appears to be in use since 
the file at [path]/.lock is locked by another process."

Is what I'm trying to do possible?  Do I have something wrong in my 
configuration that I need to check?

Thanks for your help,

Jon

Re: Repository locking with clustering

[Dominique Pfister <do...@day.com>]

Hi Jon,

On 5/8/07, Jon Walker <jw...@shokker.net> wrote:
> I've been looking at the clustering feature that was added in, but am
> having trouble with the locking.  I have 2 Model 1 deployments set up
> using clustering (each with a unique ID).  Through shared drives, they are
> accessing the same repository location (unique configurations for the
> clustering) and I have SimpleDBPersistence to SQL Server for persistence.

Sharing the repository location is not supported: every repository
must have its own, unique repository home. This is also the reason for
the exception you get.

In order to use clustering, not the actual repository homes, but the
persistence managers of your 2 deployments must point to the same
location. Furtheron, only database persistence managers currently
provide the ACID functionality required for transactionally correct
behaviour.

I recently added some wiki page to help setup clustering that might be helpful:

http://wiki.apache.org/jackrabbit/Clustering

If you have more questions or find that some important information is
missing, please do not hesitate to ask.

Kind regards
Dominique

>
> On first access to each instance, the webservices attempt to start the
> repository.  The first server to be hit will create a Repository instance
> and create the .lock file in the repository's home directory.  When the
> other server gets hit and attempts to start the repository, it throws an
> exception saying "The repository home at [path] appears to be in use since
> the file at [path]/.lock is locked by another process."
>
> Is what I'm trying to do possible?  Do I have something wrong in my
> configuration that I need to check?
>
> Thanks for your help,
>
> Jon
>