You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@felix.apache.org by Lukasz Dywicki <lu...@dywicki.pl> on 2010/05/20 16:39:21 UTC

Karaf features improvements

Hello,
I would talk about Karaf Features. At this moment features can be used to
install or uninstall set of bundles but FeaturesService does not track state
of bundles.

For example when we install servicemix-saxon feature and after that remove
saxon bundle feature state is installed but feature is not complete. When we
install bundles from some feature by hand feature should be also marked as
installed.
These two things will make Karaf Features better. :-) What do you think?

Regards,
Lukasz
-- 
View this message in context: http://old.nabble.com/Karaf-features-improvements-tp28622180p28622180.html
Sent from the Apache Felix - Dev mailing list archive at Nabble.com.


Re: Karaf features improvements

Posted by Lukasz Dywicki <lu...@dywicki.pl>.
Ok,
I will write code similar to feature state management (verify
blueprint/spring context state) and bundle monitoring based od feature sets.
I can provide patch for these things.

Regards,
Lukasz


gnodet wrote:
> 
> On Fri, May 21, 2010 at 11:42, Lukasz Dywicki <lu...@dywicki.pl> wrote:
>>
>> Hello Guillaume,
>>
>> gnodet wrote:
>>>
>>> There is some work going on in the OSGi Alliance about multiple
>>> frameworks in one JVM (see
>>> http://www.osgi.org/Download/File?url=/download/osgi-core-4.3-early-draft1.pdf
>>> for more informations).  This would enable to have a tree of OSGi
>>> frameworks instead of a single instance and would enable controlled
>>> sharing between a framework and its parent.      This would enable a
>>> controlled level of isolation and could be used to make sure some
>>> features don't interfere with each other.  That's one of the reason
>>> i'm not totally convinced about your proposal.  If we'd were to use
>>> nested frameworks, each feature would have its own framework, so that
>>> would not map well to your proposal.
>>>
>> I didn't know that OSGi alliance published new draft so fast.
>>
>> I am not familar with current draft - but I found that bundle listeners
>> will
>> be also scoped per framework - so when you'll install feature (and
>> budnles)
>> in one other instances will don't know anything about this operation.
>>
> 
> Well, the idea would be that a feature would create a new nested
> framework, so all features would be visible from the top level.
> 
>>
>>
>> gnodet wrote:
>>>
>>> The two other things i have in mind for features are:
>>>   * give a lifecycle to a feature (start / stop)
>>>   * allow more control on the bundles when installed (started level,
>>> start state)
>>>
>>>
>> I agree with you in your proposals:
>> +1 for feature lifecycle
>> +1 allow more control on the bundles when installed
>>
>>
>> gnodet wrote:
>>>
>>>   * leverage obr to compute a transitive closure of the feature before
>>> installation
>>> The last one would allow a more loose definition of a feature but just
>>> specifying the list of key bundles and let the features service
>>> determine all the required dependencies and download / install them.
> 
>> I don't know OBR and other stuff so I can't help and vote on this. I can
>> help in first two cases.
> 
> OBR is a repository containing OSGi metadata.  It would be used to
> compute the required dependencies.
> 
>>
>>
>> gnodet wrote:
>>>
>>> FWIW, nested frameworks are not available yet in felix (there's only a
>>> prototype in an equinox branch), so we can't really use those now, but
>>> the other things should be possible to implement now.  In particular,
>>> the use of OBR should now be possible since the new release as some
>>> acceptable performances.
>>
>> I think we can leave 4.3 and it's features and make Karaf good base to
>> work
>> with OSGi 4.2.
>> From my side groving osgi complexity it's bad idea - I love KISS
>> principle
>> and in my idea OSGi 4.2 is good example of simple and powerful
>> environment.
>> We should wait until 4.3 will be in final stage to get them into Karaf.
>>
>> Best regards,
>> Lukasz Dywicki
>> --
>> View this message in context:
>> http://old.nabble.com/Karaf-features-improvements-tp28622180p28631675.html
>> Sent from the Apache Felix - Dev mailing list archive at Nabble.com.
>>
>>
> 
> 
> 
> -- 
> Cheers,
> Guillaume Nodet
> ------------------------
> Blog: http://gnodet.blogspot.com/
> ------------------------
> Open Source SOA
> http://fusesource.com
> 
> 

-- 
View this message in context: http://old.nabble.com/Karaf-features-improvements-tp28622180p28654869.html
Sent from the Apache Felix - Dev mailing list archive at Nabble.com.


Re: Karaf features improvements

Posted by Guillaume Nodet <gn...@gmail.com>.
On Fri, May 21, 2010 at 11:42, Lukasz Dywicki <lu...@dywicki.pl> wrote:
>
> Hello Guillaume,
>
> gnodet wrote:
>>
>> There is some work going on in the OSGi Alliance about multiple
>> frameworks in one JVM (see
>> http://www.osgi.org/Download/File?url=/download/osgi-core-4.3-early-draft1.pdf
>> for more informations).  This would enable to have a tree of OSGi
>> frameworks instead of a single instance and would enable controlled
>> sharing between a framework and its parent.      This would enable a
>> controlled level of isolation and could be used to make sure some
>> features don't interfere with each other.  That's one of the reason
>> i'm not totally convinced about your proposal.  If we'd were to use
>> nested frameworks, each feature would have its own framework, so that
>> would not map well to your proposal.
>>
> I didn't know that OSGi alliance published new draft so fast.
>
> I am not familar with current draft - but I found that bundle listeners will
> be also scoped per framework - so when you'll install feature (and budnles)
> in one other instances will don't know anything about this operation.
>

Well, the idea would be that a feature would create a new nested
framework, so all features would be visible from the top level.

>
>
> gnodet wrote:
>>
>> The two other things i have in mind for features are:
>>   * give a lifecycle to a feature (start / stop)
>>   * allow more control on the bundles when installed (started level,
>> start state)
>>
>>
> I agree with you in your proposals:
> +1 for feature lifecycle
> +1 allow more control on the bundles when installed
>
>
> gnodet wrote:
>>
>>   * leverage obr to compute a transitive closure of the feature before
>> installation
>> The last one would allow a more loose definition of a feature but just
>> specifying the list of key bundles and let the features service
>> determine all the required dependencies and download / install them.

> I don't know OBR and other stuff so I can't help and vote on this. I can
> help in first two cases.

OBR is a repository containing OSGi metadata.  It would be used to
compute the required dependencies.

>
>
> gnodet wrote:
>>
>> FWIW, nested frameworks are not available yet in felix (there's only a
>> prototype in an equinox branch), so we can't really use those now, but
>> the other things should be possible to implement now.  In particular,
>> the use of OBR should now be possible since the new release as some
>> acceptable performances.
>
> I think we can leave 4.3 and it's features and make Karaf good base to work
> with OSGi 4.2.
> From my side groving osgi complexity it's bad idea - I love KISS principle
> and in my idea OSGi 4.2 is good example of simple and powerful environment.
> We should wait until 4.3 will be in final stage to get them into Karaf.
>
> Best regards,
> Lukasz Dywicki
> --
> View this message in context: http://old.nabble.com/Karaf-features-improvements-tp28622180p28631675.html
> Sent from the Apache Felix - Dev mailing list archive at Nabble.com.
>
>



-- 
Cheers,
Guillaume Nodet
------------------------
Blog: http://gnodet.blogspot.com/
------------------------
Open Source SOA
http://fusesource.com

Re: Karaf features improvements

Posted by Lukasz Dywicki <lu...@dywicki.pl>.
Hello Guillaume,

gnodet wrote:
> 
> There is some work going on in the OSGi Alliance about multiple
> frameworks in one JVM (see
> http://www.osgi.org/Download/File?url=/download/osgi-core-4.3-early-draft1.pdf
> for more informations).  This would enable to have a tree of OSGi
> frameworks instead of a single instance and would enable controlled
> sharing between a framework and its parent.      This would enable a
> controlled level of isolation and could be used to make sure some
> features don't interfere with each other.  That's one of the reason
> i'm not totally convinced about your proposal.  If we'd were to use
> nested frameworks, each feature would have its own framework, so that
> would not map well to your proposal.
> 
I didn't know that OSGi alliance published new draft so fast.

I am not familar with current draft - but I found that bundle listeners will
be also scoped per framework - so when you'll install feature (and budnles)
in one other instances will don't know anything about this operation.



gnodet wrote:
> 
> The two other things i have in mind for features are:
>   * give a lifecycle to a feature (start / stop)
>   * allow more control on the bundles when installed (started level,
> start state)
> 
> 
I agree with you in your proposals:
+1 for feature lifecycle
+1 allow more control on the bundles when installed


gnodet wrote:
> 
>   * leverage obr to compute a transitive closure of the feature before
> installation
> The last one would allow a more loose definition of a feature but just
> specifying the list of key bundles and let the features service
> determine all the required dependencies and download / install them.
I don't know OBR and other stuff so I can't help and vote on this. I can
help in first two cases.


gnodet wrote:
> 
> FWIW, nested frameworks are not available yet in felix (there's only a
> prototype in an equinox branch), so we can't really use those now, but
> the other things should be possible to implement now.  In particular,
> the use of OBR should now be possible since the new release as some
> acceptable performances.

I think we can leave 4.3 and it's features and make Karaf good base to work
with OSGi 4.2.
>From my side groving osgi complexity it's bad idea - I love KISS principle
and in my idea OSGi 4.2 is good example of simple and powerful environment.
We should wait until 4.3 will be in final stage to get them into Karaf.

Best regards,
Lukasz Dywicki
-- 
View this message in context: http://old.nabble.com/Karaf-features-improvements-tp28622180p28631675.html
Sent from the Apache Felix - Dev mailing list archive at Nabble.com.


Re: Karaf features improvements

Posted by Guillaume Nodet <gn...@gmail.com>.
Well, I'm not completely sure about marking a feature as installed as
soon as the  bundles are all installed, neither the opposite actually.

There is some work going on in the OSGi Alliance about multiple
frameworks in one JVM (see
http://www.osgi.org/Download/File?url=/download/osgi-core-4.3-early-draft1.pdf
for more informations).  This would enable to have a tree of OSGi
frameworks instead of a single instance and would enable controlled
sharing between a framework and its parent.      This would enable a
controlled level of isolation and could be used to make sure some
features don't interfere with each other.  That's one of the reason
i'm not totally convinced about your proposal.  If we'd were to use
nested frameworks, each feature would have its own framework, so that
would not map well to your proposal.

The two other things i have in mind for features are:
  * give a lifecycle to a feature (start / stop)
  * allow more control on the bundles when installed (started level,
start state)
  * leverage obr to compute a transitive closure of the feature before
installation

The last one would allow a more loose definition of a feature but just
specifying the list of key bundles and let the features service
determine all the required dependencies and download / install them.

FWIW, nested frameworks are not available yet in felix (there's only a
prototype in an equinox branch), so we can't really use those now, but
the other things should be possible to implement now.  In particular,
the use of OBR should now be possible since the new release as some
acceptable performances.

On Thu, May 20, 2010 at 16:39, Lukasz Dywicki <lu...@dywicki.pl> wrote:
>
> Hello,
> I would talk about Karaf Features. At this moment features can be used to
> install or uninstall set of bundles but FeaturesService does not track state
> of bundles.
>
> For example when we install servicemix-saxon feature and after that remove
> saxon bundle feature state is installed but feature is not complete. When we
> install bundles from some feature by hand feature should be also marked as
> installed.
> These two things will make Karaf Features better. :-) What do you think?
>
> Regards,
> Lukasz
> --
> View this message in context: http://old.nabble.com/Karaf-features-improvements-tp28622180p28622180.html
> Sent from the Apache Felix - Dev mailing list archive at Nabble.com.
>
>



-- 
Cheers,
Guillaume Nodet
------------------------
Blog: http://gnodet.blogspot.com/
------------------------
Open Source SOA
http://fusesource.com