aries-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alasdair Nottingham <...@apache.org>
Subject Re: JMX whiteboard project
Date Tue, 05 Jul 2011 14:09:59 GMT
Hi,

I've applied 700 and 701. The patch for 702 doesn't work for me though. I
think that is because it is a patch on 700, not 701.

Alasdair

On 5 July 2011 13:44, Felix Meschberger <fmeschbe@adobe.com> wrote:

> Hi
>
> I have created ARIES-701 [1] and ARIES-702 [2] showing where this all
> might lead to..
>
> Regards
> Felix
>
> [1] https://issues.apache.org/jira/browse/ARIES-701
> [2] https://issues.apache.org/jira/browse/ARIES-702
>
> Am Dienstag, den 05.07.2011, 12:50 +0100 schrieb Alasdair Nottingham:
> > Hi,
> >
> > This is amazing timing. I've hit exactly this problem while writing the
> code
> > for ARIES-686.
> >
> > I agree with your proposal, although when you describe 2 when you say
> you'll
> > look to see if the MBean implements a *MBean interface do you mean you
> will
> > use reflection, or look at the interfaces published into the service
> > registry? I think I'd like it to be the latter.
> >
> > Thanks
> > Alasdair
> >
> > On 5 July 2011 08:03, Felix Meschberger <fmeschbe@adobe.com> wrote:
> >
> > > Hi,
> > >
> > > I think I missed a problematic point: Standard MBeans registered with
> > > the <classname>MBean pattern require that the implementation and the
> > > MBean interface reside in the same package. Not very practical for OSGi
> > > where the MBean interface of course must be exposed.
> > >
> > > So, I think we should probably drop the requirement for the MBean to
> > > exposed with an interface matching "*MBean".
> > >
> > > I still don't like to require the MBeanRegistration interface to be
> used
> > > in the registration. It is kind of like a helper interface not
> > > identifying the primary purpose of the MBean.
> > >
> > > How about changing the filter to just be (jmx.objectname=*). If the
> > > actual property is an empty string (or is not a single-value String),
> > > the object is expected to implement the MBeanRegistration interface. If
> > > not, an ERROR level message is logged and the service ignored.
> > >
> > > Now for the actual service object being registered:
> > > 1 if the service implements the DynamicMBean interface, use it
> > >   right away as the object to register
> > > 2 otherwise see whether the object implements an interface whose
> > >   name matches the <simple-class-name>MBean pattern. If so, wrap
> > >   the object with a StandardMBean class using the interface as
> > >   the MBean interface (and log this at INFO level)
> > > 3 otherwise log an ERROR level message and ignore the service
> > >
> > > Note that using the simple-class-name in the second step is a deviation
> > > from the JMX Spec which requires the MBean interface to reside in the
> > > same package as the MBean object.
> > >
> > > To make this work, the mbeanTracker ServiceTracker must be modified to
> > > actually track all services (this is a generic bug to be fixed,
> > > ARIES-700).
> > >
> > > Regards
> > > Felix
> > >
> > > Am Montag, den 27.06.2011, 20:45 +0100 schrieb Alasdair Nottingham:
> > > > On 27 June 2011 20:19, Felix Meschberger <fmeschbe@gmail.com> wrote:
> > > >
> > > > > Hi,
> > > > >
> > > > > Am Montag, den 27.06.2011, 20:07 +0100 schrieb Alasdair Nottingham:
> > > > > > Hi,
> > > > > >
> > > > > > I've been looking at the way the whiteboard implementation works
> and
> > > I
> > > > > was
> > > > > > wondering if it would make sense to change the way it detects
> mbeans.
> > > > > > Currently it detects them by looking for:
> > > > > >
> > > > > > (objectClass=*MBean). The impl then needs to either have a
> > > jmx.objectname
> > > > > > property, or it needs to be javax.management.MBeanRegistration
> > > extension.
> > > > > I
> > > > >
> > > > > My idea for requiring some MBean interface is that it makes
> > > registration
> > > > > extremely easy:
> > > > >
> > > >
> > > > I agree with this goal.
> > > >
> > > >
> > > > >
> > > > > - either it is a DynamicMBean (or some extension thereof) service
> > > > > - or it is an interface with MBean suffix which as per the spec
> > > > >   defines the MBean interface for the bean
> > > > >
> > > >
> > > > OK.
> > > >
> > > >
> > > > >
> > > > > For the registration then only an ObjectName is required which can
> be
> > > > > provided as a service registration property or by implementing the
> > > > > MBeanRegistration interface (which is also similarly used in the
> spec
> > > > > IIRC).
> > > > >
> > > > >
> > > > > > think it would make more sense for a service filter like this:
> > > > > >
> > > > > >
> (|(objectClass=javax.management.MBeanRegistration)(jmx.objectname=))
> > > > > >
> > > > > > what do people think?
> > > > >
> > > > > By going that way, you will solve the second issue with the filter
> but
> > > > > you then have an MBean where you have to find out how to be able
to
> > > > > register (or I may be missing something in more recent JMX specs).
> > > > >
> > > > > But then, I don't think we should require the MBeanRegistration
> > > > > interface as a service interface. Sounds kind of incorrect.
> > > > >
> > > >
> > > > I think if it needs to be an MBeanRegistration then we should require
> the
> > > > object to be advertised as an MBeanRegistration. Not putting
> > > > MBeanRegistration on a service and then relying on it being one is
> dodgy
> > > in
> > > > OSGi. Sure in most cases it'll work, but if someone decides to use
> > > service
> > > > hooks to insert a proxy they will probably get this wrong, also you
> can't
> > > > make use of the service registry to get the class space consistency.
> > > >
> > > > Overall based on this I think a more correct filter would be this:
> > > >
> > > >
> > >
> (&(objectClass=*MBean)(|(objectClass=javax.management.MBeanRegistration)(jmx.objectname=)))
> > > >
> > > >
> > > > >
> > > > > All in all, I think the original filter sounds more correct.
> > > > >
> > > > > Regards
> > > > > Felix
> > > > >
> > > > > > Alasdair
> > > > > >
> > > > >
> > > > >
> > > > >
> > > >
> > > >
> > >
> > >
> > >
> >
> >
>
>
>


-- 
Alasdair Nottingham
not@apache.org

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message