[muscle] SV: SV: SV: Re: SV: Re: Server Design issue.

Actually the EndServer( ) call doesn't work so nice that I'd thought.

What I do is that I call the following in my MessageReceivedFromGateway in
my derivied StorageReflectSession class (a "server-stopper-client" sends the
NB_COMMAND_STOP_SERVER message to the server):

        switch ( msg->what )
        {
                case NB_COMMAND_STOP_SERVER:
                {
                  Cleanup( );
                  EndServer( );
                  break;
                }
        ...

If I'm not calling Cleanup( ) MUSCLE will be very angry and say that I
probably hasn't called Cleanup( ). This will result in a lot of LogTime's
and a segmentation fault. So I added Cleanup( ) ...but...

On RH9 (linux,g++,pthreads) the application exits with the "Killed" message
to the std output. It is strange - it is just if I had issued "kill -9 pid"!

On Windows (msvc,win32) the application crashes with "Unhandled
exception..." at line 1007 in storagereflectsession.cpp.

What am I missing in order to stop the server correctly?

/D

-----Opprinnelig melding-----
Fra: muscle-bounce@xxxxxxxxxxxxx [mailto:muscle-bounce@xxxxxxxxxxxxx]På;
vegne av David Svanberg
Sendt: 27. oktober 2003 10:11
Til: muscle@xxxxxxxxxxxxx
Emne: [muscle] SV: SV: Re: SV: Re: Server Design issue.


Sorry,

next time I should consult the docs first. EndServer seems to be a pretty
nice little method to achieve want I want...

...sorry...

/D

-----Opprinnelig melding-----
Fra: muscle-bounce@xxxxxxxxxxxxx [mailto:muscle-bounce@xxxxxxxxxxxxx]På;
vegne av David Svanberg
Sendt: 27. oktober 2003 10:06
Til: muscle@xxxxxxxxxxxxx
Emne: [muscle] SV: Re: SV: Re: Server Design issue.


Hi,

is there a way to stop the server from a StorageReflectSession object?

I have a client sending a NB_COMMAND_STOP_SERVER message and wants actually
to stop the server when the server gets this message... :)

/D

-----Opprinnelig melding-----
Fra: muscle-bounce@xxxxxxxxxxxxx [mailto:muscle-bounce@xxxxxxxxxxxxx]På;
vegne av Jeremy Friesner
Sendt: 22. oktober 2003 17:54
Til: muscle@xxxxxxxxxxxxx
Emne: [muscle] Re: SV: Re: Server Design issue.


> It should be good if one could do a ServerProcessSingleLoop that runs one
> iteration through the ServerProcessLoop leaving control back to the
caller.

That wouldn't work very well, because each iteration through the loop
contains a call to select(), and select() may block for a long period of
time
(it willl block until the next piece of I/O is ready for processing, or
until
the next PulseNode event-time has arrived).

> Having the possibility to install a callback function from the
> ServerProcessLoop giving us the chance to interact.

You can get lots of callback hooks by installing a Session object and
overriding some of its methods to do what you want.  (The session
object doesn't have to be connected to a client... you can pass in
socket=3D-1 to indicate there is no connection)

Jeremy

Other related posts: