It might not fit too well with the felix/karaf idea of installing config admin 
pid configs through .cfg property files, but config admin has no problem 
dealing with a string property that is an entire xml document.  So I'd go for a 
handy way to initialize such config admin properties rather than a whole new 
logback configurator.  Maybe some kind of notation like

..file.key=filelocation

which means the value for key is read from the filelocation.

thanks
david jencks

On Oct 14, 2011, at 9:33 AM, Guillaume Nodet wrote:

> 
> 
> On Fri, Oct 14, 2011 at 18:27, ceki <[email protected]> wrote:
> On 14/10/2011 5:48 PM, Guillaume Nodet wrote:
> 
> Thank you for your quick response.
> 
> 
> I do care a lot about logback being configure with properties to be able
> to leverage ConfigAdmin.  It should be *the* way to configure things in
> OSGi.  That way, you can distribute the configuration remotely or store
> it in a DB or in any other mean without having to rewrite all the
> bundles to leverage that.  That's the benefit of using a standard service.
> 
> How does any of the above change for properties format. For log4j which 
> supports properties format, you still need to invoke PropertyConfigurator on 
> the properties (or some URL containing the properties). It would be no 
> different with logback, except that you would invoke a different configurator.
> 
> 
> Yeah, I agree.  I just don't want to write that configurator.
>  
> 
> 
> You just said the configuration file needs to be xml or groovy, which is
> different from a properties file.  For config admin, the input data
> needs to be a map of key/value pairs.  I haven't said it was not
> possible with logback, just that it does not exist, and I don't have the
> time and will to start writing a new configuration mechanism for logback
> without having any real need to switch to it.
> 
> But if you want to try that, it could be nice.  Though I still haven't
> heard the reasons why you want logback instead of pax-logging.
> 
> I am not a Karaf user, at least not yet. I am the founder of both log4j and 
> logback projects although I now work mostly on logback. I am just trying to 
> understand your use case for properties configuration. My apologies if the 
> use case is obvious for Karaf users.
> 
> The use case is to provide the configuration through the standard OSGi 
> ConfigAdmin service.  The main benefit is that the data storage can be 
> abstracted.  We use files by default as the primary source, but Cellar can 
> push configuration changes between instances, Fuse Fabric stores them in 
> Zookeeper, I've also seen people using a JDBC storage for the configuration.  
> All the things are mostly interesting when dealing with large deployments, 
> not for a single instance, I agree.
>  
> 
> Best regards,
> --
> Ceki
> 
> 
> 
> -- 
> ------------------------
> Guillaume Nodet
> ------------------------
> Blog: http://gnodet.blogspot.com/
> ------------------------
> Open Source SOA
> http://fusesource.com
> 

Reply via email to