I was not aware we have this information in the wires. I will try to
implement it using these.
So what I would do is this:
I start with the features to be stopped / started.
I then propagate down to the subfeatures / bundles.
For each feature processed I compute the state as the highest state of
all features that depend on it and the requested state.
Does that make sense?
The problem in this algorithm is that in many cases we will not be able
to stop features / bundles as they are needed by other features.
So probably we will need some reporting to explain why some features /
bundles could not achieve the request state. To really stop a feature
the user would then
have to stop all top level features. Only then would it really change
its state.
We could also have kind of a force mode where we change the state first
to the top and then down again. So this would guarantee that a feature
changes its state but it could shut down half of karaf unintendedly.
Which would be especially dangerous if we hit some core features like
the feature service. So I tend to rather not support this.
Christian
On 13.04.2015 18:01, Guillaume Nodet wrote:
Yeah, I was thinking about it.
Though the obvious other solution is to fix it.
I have actually started an email this morning to discuss but I haven't
finished it.
Overall, I think it may not be very difficult to fix, as the bundle state
changes are already handled correctly afaik. The real problem is to agree
on the semantics on the effects, so that we can compute the desired state
of each bundle correctly.
Problems arise when a bundle is used by several features, one of which
being started and the other resolved.
Anyway, it's really up to you, I don't mind fixing the code as long as we
agree on the behaviour.
2015-04-13 17:51 GMT+02:00 Jean-Baptiste Onofré <[email protected]>:
Hi all,
I discussed with Christian about KARAF-3102.
The feature lifecycle doesn't actually fully work, especially around the
stop action.
In order to avoid to perturb the users, I think we should remove the
features lifecycle commands. Else, if they are provided, users will try it
and may be disappointed as they won't work as expected.
WDYT ?
Regards
JB
--
Jean-Baptiste Onofré
[email protected]
http://blog.nanthrax.net
Talend - http://www.talend.com
--
Christian Schneider
http://www.liquid-reality.de
Open Source Architect
http://www.talend.com
