| Re: [ecf-dev] Egg cooking |
|
Hi Wim,
You might also want to make the following point wrt remote services operating on unreliable networks: 1) OSGi Remote Services spec *allows* implementations to map network failure (e.g. network partitions, host failures, etc) onto OSGi service dynamics. What do I mean by this? Well, let's say you are a client/consumer of a remote service, and the network partitions/fails underneath that remote service. One intuitive possibility is that for the consumer, that the OSGi remote service proxy goes away (gets unregistered locally). This can/would trigger things like DS unbind/bind, ServiceTracker notification, and/or ServiceListener notification on the consumer of the remote service, allowing it to respond at runtime. For example, in your hotplate scenario...there could be logic for shutting down/off the hotplate if the remote service disappears because of network partition or host failure. 2) To get the runtime behavior described in 1, however, it depends upon some sort of reasonably timely *failure detection* in the RS/RSA implementation. Such failure detection is not at all guaranteed by the RS specification, and in fact, is only even known by me to be present in ECF's RS implementation. 3) With ECF's provider architecture, it's quite possible to design/develop/test the remote service using one (or several) remote service providers, and then test/deploy with either the same or some other provider. This makes it quite easy for failure detection to be *added or changed* as required by the application needs, without totally replacing the RS/RSA implementation. Also, with some remote services the non-functional requirements for reliability may very well change during impl/testing/deployment, and this can be easily accommodated without having to replace the entire RS/RSA implementation. This flexibility is one very tangible benefit of ECF's open, mature APIs, modular provider structure. To be direct about it, I don't know of any other RS/RSA impl where such modularity even exists. Sorry for the long posting, but reliability/partial failure is often a very big deal for applications (sometimes for safety reasons as in your demo!)...and it's very difficult to handle in general for distributed systems. So it might be helpful if we made these points about ECF's open modularity/provider architecture as frequently and as publicly as possible. Thanks, Scott On 8/16/2014 3:59 AM, Wim Jongman wrote:
|