Date: Wed, 22 Apr 2009 10:18:25 -0700
From: Scott Lewis <
slewis@xxxxxxxxxxxxxxxxx>
Subject: [ecf-dev] handling pub-sub providers with RFC 119
To: "Eclipse Communication Framework (ECF) developer mailing list."
<
ecf-dev@xxxxxxxxxxx>
Message-ID: <
49EF5161.2020405@xxxxxxxxxxxxxxxxx>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Hi Folks,
Recently, there was a message thread titled "Remote service with
ActiveMQ JMS topic using RFC 119 transparent way. " and in this
message
[0], I described some work I'm doing with the ECF RFC 119 impl to easily
deal with these cases (i.e. option 2 in my message). By 'these cases' I
mean publish and subscribe groups that are used for remote services
between members of those groups (e.g. JMS groups, etc).
Currently, there's no direct support in RFC119 for remote services
within pubsub groups...because although RFC119 has the notion of an
'endpoint ID', this endpoint ID gives the location of the remote
service's host...which assumes that the host is a server (i.e. can be
reached *directly* from a client), rather than indirectly (i.e. via a
pub/sub group).
ECF's IContainer can represent communication with either an endpoint
(server) *or* a pubsub group and this makes it possible to use this
capability to support peer-to-peer remote services within a pub-sub
group (e.g. a jms topic).
In order to make this
capability accessible via the ECF RFC119 impl
(please add to your thoughts about giving a name to ECF's impl here [1])
, I've added some API to our RFC119 implementation in
org.eclipse.ecf.osgi.services.distribution. What it does is expose both
the *endpointID* and the *connectTargetID* in the discovery-published
meta-data (i.e. see this interface [2]). On the client side, this
information about the remote service (both it's endpointID and
connectTargetID) can then be used to connect to a pub/sub group at
remote service discovery time (but only when the connectTargetID is
non-null...note that this has *no* effect/relevance for client-server
distribution providers).
I've implemented the mechanism to do this, and am working on some test
code using the ActiveMQ (JMS) provider. If others are interested in
helping test this, please let me know.
One more thing...in this previous
thread [3] I brought up the issue of
dealing with the situation where potentially many remote services were
published/available on a lan (e.g.) and it was unclear (from RFC119) how
this should be handled...e.g. does the announcement of some remote
service mean that another remote service doesn't get a proxy? Should
all remote service discoveries lead to proxy creation? How can an
application provide some control/scope to the proxy creation?
RFC119 doesn't really address this situation yet, but I've added two
service interface called IProxyContainerFinder, and IHostContainerFinder
[4] that allows other bundles to register themselves as 'container
finders'...i.e. to decide which (of potentially many) remote services
providers should handle a discovered service publication. Note that
these container finders allow full customization of the remote service
discovered meta-data
(i.e. they can decide to create new IContainers,
reuse old ones, ignore some, etc). The way to replace the
DefaultProxyContainerFinder is to simply register your own
implementation of IProxyContainerFinder (or IHostContainerFinder) as an
OSGi service...e.g:
bundleContext.registerService(IProxyContainerFinder.class.getName(),new
MyProxyContainerFinder(), null);
And when services are discovered via the discovery impl(s) your
container finder will be consulted to make the decision about what
IContainer should/will handle the request. Note that currently the
service with the highest service ranking will be selected (if there are
many), and the DefaultProxyContainerFinder has the lowest service
ranking possible (so will only be used if *no* other container finders
are registered in the service registry).
Note that there is also a IHostContainerFinder, to allow the same sort
of
customization for selection of the IContainers to handle the host
service publication (with the default being all existing IContainers).
Any comments welcomed. I'm working on testing/debugging this API right
now (it already works fine with the r-OSGi and ecf generic providers
(i.e. in org.eclipse.ecf.tests.osgi.services.distribution), but I want
to verify the correct behavior for ActiveMQ provider also, as above).
To summarize, the two API additions are:
1) IServiceEndpointDescription including both endpointID and connectTargetID
2) Adding IProxy/HostContainerFinder API to allow for
customization/extension of the default container selection mechanism for
both host and proxy handling.
Thanks,
Scott
[0]
http://dev.eclipse.org/mhonarc/lists/ecf-dev/msg02265.html[1]
https://bugs.eclipse.org/bugs/show_bug.cgi?id=270652[2]
http://www.eclipse.org/ecf/org.eclipse.ecf.docs/api/org/eclipse/ecf/osgi/services/discovery/IServiceEndpointDescription.html[3]
http://dev.eclipse.org/mhonarc/lists/ecf-dev/msg02206.html[4]
http://www.eclipse.org/ecf/org.eclipse.ecf.docs/api/org/eclipse/ecf/osgi/services/distribution/package-summary.html