RE: FW: Bug in Jonas/Jeremie - Increasing Thread count

[Halas, Miroslav]

Thanks Bruno for your time and help. This is a great news. Do you have any expectation when you expect new binary release of Jonas (this is quite big issue for us)

 

Miro

-----Original Message-----
From: Bruno Dumant [mailto:[EMAIL PROTECTED]]
Sent: Monday, December 18, 2000 10:22 AM
To: Halas, Miroslav
Cc: DUMANT Bruno FTRD/DTL; '[EMAIL PROTECTED]'; Jonathan
Subject: Re: FW: Bug in Jonas/Jeremie - Increasing Thread count

 
Miro,

I have finally reproduced your problem, thanks to the traces your sent me last friday.

As I said before, the problem was not with threads but with connections, and more precisely with the handling of localhost. It seems that you try to access your registry using localhost as host name. I was able to reproduce your problem using the -Djava.naming.provider.url=jrmi://localhost option. Then, there were two problems:

• first, the registry was not recognised as being local. This is due to the fact that to recognise that a server is local, Jonathan tries to match the host address and port number of the server with those of the created server sockets. When the network is on, the host address of the server socket is the real IP address, not 127.0.0.1.
  Since the registry was not recognised as being local, Jonathan tries to create a client connection to it.
• second, client connections were never reused. This is due to a similar reason : client connections are stored in a hashtable, indexed by their host address and port number, the host address used being the real host name of the socket. Since, when the network is on, the host address is never 127.0.0.1, the already present client connections were not reused, and a new one was created, causing the creation of two threads : one for the client side, the other for the server side.

The problem is now solved. I have committed the changes in the CVS repository. They should be available in the next Jonas release.

"Halas, Miroslav" wrote:

Then the next step was that I made the naming context static so all beans
are reusing the same naming context. The number of created threads decreased
a lot (we still create NamingContext for each HTTP request in the servlets)
which signals me that the threads are connected as a consequence of creation
of naming context and these threads are never terminated when the naming
context is closed.

Yes,  a new registry reference was created for each naming context, and consequently a socket connection and two threads.

My questions are:

1. Is it correct what we do with naming context (to keep it open for longer
time)?

I believe it is...

2. Is there a connection between creating naming context in the beans and
creation of threads in Jonas JVM?

There was. Now, even if the registry is remote, you will be able to reuse an existing connection, which means that new threads won't be created.

3. Is is a bug in Jonas/Jeremie that these threads are not cleand up when we
close the naming context or is it our problem?

No, it was not the problem. The problem was that created threads were not closed when the reference to the registry was lost. This is also solved now.

Thanks to everybody for help with diagnosing this problem,

Thanks for your patience !

Bruno
 

--
*******************************************************************
Bruno Dumant
DTL/ASR
france telecom R&D
38-40 rue du g�n�ral Leclerc
92794 Issy Moulineaux Cedex 9
FRANCE

mailto:[EMAIL PROTECTED]
tel: +33 1 45 29 46 03
fax: +33 1 45 29 66 04
*******************************************************************