river-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter Jones <...@roundroom.net>
Subject Re: strange problem with DGC
Date Sun, 07 Mar 2010 14:36:03 GMT
What transport layer implementation are you using?

Here is the way that this case is supposed to be covered:

- The OutputStream returned by InboundRequest.getResponseOutputStream on the server side implements
net.jini.io.ObjectStreamContext, which contains an element that implements net.jini.io.context.AcknowledgmentSource.

- BasicObjectEndpoint.writeObject (which represents the "live references" in the DGC sense)
detects this element during marshalling and registers an AcknowledgmentSource.Listener with
it, which keeps the BasicObjectEndpoint strongly referenced until its acknowledgmentReceived
method is invoked.

- When its acknowledgmentReceived method is invoked, either with true to indicate that the
reading side has successfully unmarshalled the data (and thus registered the BasicObjectEndpoint
with DGC) or with false to indicate that the transport layer has failed or given up, then
the listener drops its strong reference.

The standard net.jini.jeri.* transport layer implementations should implement this AcknowledgmentSource
protocol, but I'm wondering if you are using some other implementation that doesn't, or a
layered implementation that needs to expose the underlying functionality.

I'm surprised (or I should say embarrassed) to see that the relationship between DGC and this
AcknowledgmentSource protocol is not clearly called out in the net.jini.jeri package spec--
the mechanism only seems to be covered in the serialized form spec for BasicObjectEndpoint.writeObject.

FYI, this return value case was well anticipated (back to [1]), it's actually the parameter
case that can be surprisingly vulnerable to modern JVM GC improvements:


although that bug was for JRMP, and JERI's BasicInvocationHandler is incidentally not susceptible
to that problem because of the OutboundRequestIterator.

-- Peter

[1] http://www.hpl.hp.com/techreports/Compaq-DEC/SRC-RR-116.pdf

On Mar 5, 2010, at 5:51 AM, Sim IJskes - QCG wrote:

> Peter Firmstone wrote:
>> Note that when you export your service it's proxy is stored in Marshalled form until
the client discovers it, while in Marshalled form 
> Eh, that isn't exactly true is it? After export, you get a Proxy with a reference to
a BasicInvocationHandler with a reference to a BasicObjectEndpoint with a reference to a ImplContainer
with a reference to your service.
> As long as you keep the reference to the exported service, the service is 'strongly reachable'
from the exported service.
> The problem starts when you marshall it in BasicInvocationDispatcher.marshalReturn(),
and the reply has enough latency to arive 'late' at the client side.
> In this window the service becomes 'weakly reachable' and if the GC kicks in before DgcServer.dirty
the object gets finalized.
> Gr. Sim
> -- 
> QCG, Software voor het MKB, 071-5890970, http://www.qcg.nl
> Quality Consultancy Group b.v., Leiderdorp, Kvk Leiden: 28088397

View raw message