felix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Julian Sedding (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FELIX-3360) Bundle location is statically set for dynamically bound bundles
Date Wed, 22 Feb 2012 10:04:48 GMT

    [ https://issues.apache.org/jira/browse/FELIX-3360?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13213502#comment-13213502

Julian Sedding commented on FELIX-3360:

> The spec states that a configuration retrieved with the single argument getConfiguration(String
> method must be statically bound to the calling bundle. So everything works correctly.

My understanding is that the bundle location is indeed bound statically when a call to getConfiguration(String
pid) *creates* the configuration. If the configuration *already exists* and is not yet bound
(i.e. bundle location is null), it is then dynamically bound to the calling bundle, and unbound
when this bundle gets uninstalled.

If you don't agree, I would be grateful, if you could point me to the relevant section(s)
in the spec. The rest of the argument is based on the assumption that the above understanding
is correct.

The scenario I observed while debugging the issue is, of course, a little more complex:
1. Sling's OSGi Installer (it's what the spec refers to as "management bundle") creates the
configuration (IMO correctly) using getConfiguration(String pid, null). [0]
2. Somewhere (I didn't observe where exactly) an event of type ConfigurationEvent.CM_UPDATED
is fired.
3.The Event is handled o.a.f.scr.impl.config.ConfigurationComponentRegistry#configurationEvent,
which calls #getConfiguration and finally ConfigurationAdmin#getConfiguration(String pid).

In the issue description I was referring to the single argument call to getConfiguration in
step three. I.e. a previously unbound Configuration object gets bound dynamically to the bundle,
which retrieves the existing Configuration. In this case, I expect the Configuration to get
unbound, when the bundle is uninstalled. See also [1].

The changes in the patch reflect this scenario. A new Configuration object, created via one
of the single argument methods, should have its staticBundleLocation field already set upon
construction. So the point where the dynamicBundleLocation gets set can never be reached (it
is guarded by a null check for bundle location).

OSGi Enterprise Spec Release 4, Version 4.2, Configuration Admin Service Spec Version 1.3:
[0] 104.4.1 Location Binding: "These methods [the two argument versions of getConfiguration
and createFactoryConfiguration] are intended for management bundles."
[1] 104.4.1 Location Binding: "A null location parameter may be used to create Configuration
objects that are not bound. In this case, the objects become bound to a specific location
the first time that they are used by a bundle. When this dynamically bound bundle is subsequently
uninstalled, the Configuration object’s bundle location must be set to null again so it
can be bound again later."
> Bundle location is statically set for dynamically bound bundles
> ---------------------------------------------------------------
>                 Key: FELIX-3360
>                 URL: https://issues.apache.org/jira/browse/FELIX-3360
>             Project: Felix
>          Issue Type: Bug
>          Components: Configuration Admin
>    Affects Versions:  configadmin-1.2.8
>            Reporter: Julian Sedding
>         Attachments: FELIX-3360.patch
> The method ConfigurationAdminImpl#getConfiguration(pid) dynamically sets the configuration's
bundle location if it is null. However, it uses ConfigurationImpl#setStaticBundleLocation(location)
in order to do so. This should be changed to set the dynamic location. Attached is a proposed
> The issue I observed, that lead me to this bug, was as follows:
> 1. Installed a number of factory configurations and the bundle (version 1.0.0, bundle
location: jcrinstall:/apps/sample/install/bundle-1.0.0.jar) providing the service implementation
(using SCR) via Sling's OSGi Installer
> 2. Uninstalled the bundle.
> 3. Installed the bundle (version 1.0.2, bundle location: jcrinstall:/apps/sample/install/bundle-1.0.2.jar)
> After this, the factory configurations were not bound to the bundle any longer, because
they were still bound to the no longer existing bundle location jcrinstall:/apps/sample/install/bundle-1.0.0.jar.
This basically leaves stale configurations and a badly configured system.
> While Sling's OSGi Installer handles updates without changing the bundle location normally,
the above scenario differs in that instead of an update, there is an uninstall + re-install
happening. The Configuration Admin Service Specification 1.3 clearly states (in 104.4.1 Location
Binding): "When this dynamically bound bundle is subsequently uninstalled, the Configuration
object's bundle location must be set to null again so it can be bound again later."
> Note: in the patch I also changed the return type of ConfigurationImpl#tryBindLocation()
from boolean to void. Before, it always returned true, so the return value is meaningless.
I almost ended up using it in an if statement because of the return type, which made me believe
that it returns true if the bundle location is set and false otherwise.

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira


View raw message