On 04.02.2012 14:20, Marcel Offermans wrote: > On Feb 4, 2012, at 12:41 PM, Holger Hoffstätte wrote: >> On 03.02.2012 15:49, Richard S. Hall wrote: >>> I also know that the ECF guys ran into this same issue (with Equinox) >>> and that they solved it by introducing a separate proxy bundle under >>> which they'd register their service factories. >> >> This is what I did in my (more precisely Paremus') implementation as well >> and it worked right away. Creating the synthetic bundle is easy enough, >> and since it can be created on the fly you can easily remove it in tow >> when e.g the distribution provider is uninstalled. >> The main point of the proxy bundle is to avoid polluting the provider >> bundle itself with all sorts of per-client wirings. > > Out of curiosity, how does this scale when you have to publish hundreds > or even thousands of services? The overhead of having a whole bundle > being installed, resolved and started sounds big compared to just > registering a service.
Not sure I understand the problem. The proxy bundle is only generated & installed once (by the provider for the provider) and only serves as conduit for class space wiring; you don't have a proxy bundle for every imported or exported service. That would really be overkill and pointless. It has no (measurable) impact whatsoever on the number of exported or imported services. That's mostly a consequence of the scalability of your distribution provider, discovery scoping and visibility management by the topology manager. -h --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@felix.apache.org For additional commands, e-mail: users-h...@felix.apache.org
