Philippe Durieux said on 28-11-00
(see http://www.objectweb.org/messages/JonasUsers/2000/11/msg00144.html)
>>This is not normal if your orb is rmi, because the garbage collector
>>should remove the Remote objects. All is normal if your orb is Jeremie
>>because the garbage collector is not distributed.
>>You can try to run the garbage collector in the server (JonasAdmin -gc)
>>to see if the objects are removed then.
So it would seem that the first possibility you mentioned is the case.
My situation is, I have a project soon to go live. I'd rather use Jeremie
because my EJBs make a *lot* of intercomponent calls, so I get a big
performance boost from it. But if the memory leakage can't be fixed, I'll
have to revert to RMI (which now supports security - thank you John Ellis!)
Joe
-----Original Message-----
From: Bruno Dumant [SMTP:[EMAIL PROTECTED]]
Sent: Thursday, December 07, 2000 8:49 AM
To: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Subject: Re: JOnAS 2-2-2 is available
Hi,
I have made new tests of Jeremie to see if I could reproduce the memory
leak, but I couldn't.
The test is simple :
// simplified client code
Tester tester = (Tester) Naming.lookup("initial-server-name");
while (true) {
String new_name = tester.newName();
tester = (Tester) Naming.lookup("initial-server-name");
}
// simplified server code
class TesterImpl implements Tester {
static int id = 0;
String newName() {
String oldname = "tester" + i;
Naming.unbind(oldname);
i++;
String new_name = "tester" + i;
Naming.rebind(new_name, this);
return new_name;
}
}
The client thus repeatedly invokes the server that calls the registry to
add and remove objects. This test has been running all night without any
problem. I have also tested a slightly different scenario in which a new
TesterImpl object is put in the registry during each iteration, the old
one being unexported (this scenario requires methods I have just added
to Jeremie). It works equally well.
This shows that the problem is not with the registry implementation, nor
with a memory leak during Jeremie invocations. I can think of two other
possibilities:
* the problematic examples require distributed garbage collection,
provided by RMI but not by Jeremie (and not by RMI/IIOP);
* there is a memory leak in the transaction context transmission
(not tested by the above example).
That's all for now !
Best regards,
Bruno
Philippe Coq wrote:
Joe Gittings wrote:
>
> Philippe,
>
> Thanks for this new release.
>
> Since I don't want to take credit for work I didn't do, I should
perhaps
> clarify: my contribution to security was identifying and proposing
fixes to
> a couple of bugs in the JRMI security implementation. I did not do
anything
> to extend security to work with RMI (although I guess the bugs would
have
> affected that too :).
>
> One question about 2.2.2: does it address the memory leakage problems
that
> various people (including myself) are experiencing with Jeremie under
> 2.1.1?
>
> With thanks,
> Joe
>
In fact, I must confess that I don't know exactly...
I was so involved in the release process of the version that
I have not tried to reproduce the "memory leakage problems with Jeremie"
If somebody can easily reproduce it and tell us if the new version
of jonathan fix the bug it will be fine!
Best regards,
--
Philippe
Philippe Coq Evidian Phone: (33) 04 76 29 78 49
Bull S.A - 1 rue de Provence - 38432 Echirolles Cedex France
Download our EJBServer at http://www.objectweb.org
<http://www.objectweb.org>
----
To unsubscribe, send email to [EMAIL PROTECTED] and
include in the body of the message "unsubscribe jonas-users".
For general help, send email to [EMAIL PROTECTED] and
include in the body of the message "help".
--
*******************************************************************
Bruno Dumant
DTL/ASR
france telecom R&D
38-40 rue du general Leclerc
92794 Issy Moulineaux Cedex 9
FRANCE
mailto:[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>
tel: +33 1 45 29 46 03
fax: +33 1 45 29 66 04
*******************************************************************
<< File: ATT00000.html >>
