Mailing-List: contact issues-help@karaf.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@karaf.apache.org
Date: Sat, 18 Jun 2016 17:29:05 +0000 (UTC)
From: "Benjamin Graf (JIRA)" <jira@apache.org>
To: issues@karaf.apache.org
Message-ID: <JIRA.12980566.1466261731000.34896.1466270945304@Atlassian.JIRA>
In-Reply-To: <JIRA.12980566.1466261731000@Atlassian.JIRA>
References: <JIRA.12980566.1466261731000@Atlassian.JIRA> <JIRA.12980566.1466261731000@arcas>
Subject: [jira] [Comment Edited] (KARAF-4578) Fileinstaller does not install
 bundle to root region
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
archived-at: Sat, 18 Jun 2016 17:29:07 -0000


    [ https://issues.apache.org/jira/browse/KARAF-4578?page=3Dcom.atlassian=
.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=3D1533=
8060#comment-15338060 ]=20

Benjamin Graf edited comment on KARAF-4578 at 6/18/16 5:28 PM:
---------------------------------------------------------------

I do deploy a simple bundle with a little blueprint namespace handler via f=
ile installer. The blueprint dependency is resolved via feature (standard S=
MX 7.0m2 installation). The namespace handler bundle does not get resolved =
because of missing requirement osgi.wiring.package; (&(osgi.wiring.package=
=3Dorg.apache.aries.blueprint)(version>=3D1.5.0)(!(version>=3D2.0.0))). I d=
id some debugging and recognized that the bundle is not installed in the ro=
ot region and therefor does not see the resolved blueprint core bundle. Fur=
ther investigation shows that org.apache.karaf.features.core does its activ=
ation in a separate thread (karaf-util BaseActivator class architecture). T=
herefor the feature deploy process and the file installer process are worki=
ng parallel since the framework start-level goes immediately to 100. I thin=
k that there is no synchronization while installing the eclipse region hook=
 which puts the already installed bundles into root region. If I stop (debu=
g breakpoint) the activation of org.apache.karaf.features.core and let it r=
un after all bundles have been installed by file installer everything is fi=
ne. That's why I think its a multi-thread race condition.


was (Author: graben):
I do deploy a simple bundle with a little blueprint namespace handler. The =
blueprint dependency is resolved via feature (standard SMX 7.0m2 installati=
on). The namespace handler bundle does not get resolved because of missing =
requirement osgi.wiring.package; (&(osgi.wiring.package=3Dorg.apache.aries.=
blueprint)(version>=3D1.5.0)(!(version>=3D2.0.0))). I did some debugging an=
d recognized that the bundle is not installed in the root region and theref=
or does not see the resolved blueprint core bundle. Further investigation s=
hows that org.apache.karaf.features.core does its activation in a separate =
thread (karaf-util BaseActivator class architecture). Therefor the feature =
deploy process and the file installer process are working parallel since th=
e framework start-level goes immediately to 100. I think that there is no s=
ynchronization while installing the eclipse region hook which puts the alre=
ady installed bundles into root region. If I stop (debug breakpoint) the ac=
tivation of org.apache.karaf.features.core and let it run after all bundles=
 have been installed by file installer everything is fine. That's why I thi=
nk its a multi-thread race condition.

> Fileinstaller does not install bundle to root region
> ----------------------------------------------------
>
>                 Key: KARAF-4578
>                 URL: https://issues.apache.org/jira/browse/KARAF-4578
>             Project: Karaf
>          Issue Type: Bug
>          Components: karaf-core, karaf-feature
>    Affects Versions: 4.0.5
>            Reporter: Benjamin Graf
>            Assignee: Jean-Baptiste Onofr=C3=A9
>            Priority: Critical
>
> The file installer is starting to work in parallel to the activation of o=
rg.apache.karaf.features.core which does the region hook initialization. Th=
is causes that bundles are installed to no region causing them not to get s=
ucessfully resolved. Might also be the cause of KARAF-4498.


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)