jackrabbit-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Philipp Salzgeber" <astrogr...@gmx.net>
Subject Re: jackrabbit clustering
Date Tue, 14 Aug 2007 13:26:43 GMT

-------- Original-Nachricht --------
Datum: Tue, 14 Aug 2007 09:06:27 +0200
Von: "Dominique Pfister" <dominique.pfister@day.com>
An: users@jackrabbit.apache.org
Betreff: Re: jackrabbit clustering

> Hi Philipp,
> 
> On 13/08/07, astrograph@gmx.net <astrograph@gmx.net> wrote:
> > hi,
> >
> > we have a repository which contains a lot of big binary data, we want to
> use clustering, and have the following questions:
> >
> > - Is there a way to configure the db connection used for the journal in
> a different way than to give the jdbc connection parameters? A jndi data
> source would be more useful to us.
> 
> Not yet. If you'd like to see this feature, please file a JIRA issue.

We might do that, jackrabbit is deployed as a resource adapter in an ear file, that makes
changing values in the repository.xml file for different deployments very painful.
> 
> > - With the clustered configuration we are limited to a db persistence
> manager, with the kind of data we ingest into our system we see a huge
> increase in the time it takes to process them. I think this is caused by the db
> persistence manager - do we REALLY have to use a db persistence manager?
> 
> I'm afraid so, yes. File-based persistence managers do not provide
> transaction isolation, e.g. one cluster node could potentially
> overwrite elements in the file system before the whole operation is
> actually committed.

ouch, with really big binaries (100s of MB) our dbas will find this a little troubling...

> 
> > - we found that you donĀ“t need to give a cluster id, it seems there is
> an algorithm to provide individual servers with cluster ids - is this safe
> to use?
> 
> This algorithm assigns a random cluster id to a cluster node: I'd
> rather use a well-defined, sensible cluster id, which also helps when
> tracking down what cluster node was responsible for a change.

yes, we have seen that it is possible to set this via a system property.

> 
> > - versioning is still configured for the local file system, is this ok
> for a clustered installation?
> 
> No, again you'll need a db persistence manager.

we changed that!

> 
> Kind regards
> Dominique

thx a lot

Philipp

-- 
GMX FreeMail: 1 GB Postfach, 5 E-Mail-Adressen, 10 Free SMS.
Alle Infos und kostenlose Anmeldung: http://www.gmx.net/de/go/freemail

Mime
View raw message