Newsgroups: comp.parallel.pvm
From: woyna@eid.anl.gov (Mark Woyna)
Subject: Hanging on pvm_exit() and "Zombie" tasks
Organization: Argonne National Laboratory
Date: 19 Mar 1996 21:19:50 GMT
Message-ID: <4in8dm$crd@atlantis.dis.anl.gov>


We are currently running PVM with ParcPlace-Digitalk's VisualWorks 2.0
Smalltalk product on Solaris 2.4. We have created shared libraries
(libpvm3.so and libpvm3s.so) to utilize PVM with VisualWorks DLL and C
Connect product which allows Smalltalk to call C functions. I believe we
are running PVM version 3.3.7.

Everything has been working fine until we recently began porting our
VisualWorks code to version 2.5. We are now experiencing a strange
behavior which causing our application to hang. The only change is that
the Visualworks executable, the virtual machine, is different between the
two releases. The PVM libraries have not changed.

After starting pvm and making the first call to pvm_mytid(), we get the
following on the console:

pvm> ps
                    HOST      TID   FLAG 0x COMMAND
                 jericho        0         0 -           
                 jericho    40002       4/c -

The is some form of "zombie" task that is being created which did
not appear before.

After the application calls pvm_exit(), both tasks disappear as expected:

pvm> ps
                    HOST      TID   FLAG 0x COMMAND

If we then run the application again and perform another pvm_mytid(),
the "zombie" disappears:

pvm> ps
                    HOST      TID   FLAG 0x COMMAND  
                 jericho    40003       4/c -

However, when we try to call pvm_exit(), the application hangs and
we have to kill the pvm daemon.

Does anyone have any idea what these "zombie" tasks might be and
whether or not they may be effecting the daemon?

Thanks for any input.

 
===============================================================================
Mark A. Woyna                            |
Section Manager - Information Systems    | "No free man shall ever be debarred
Advanced Computer Applications Group     |  the use of arms. The strongest
Decision & Information Sciences Division |  reason for the people to retain the
Argonne National Laboratory              |  right to keep and bear arms is,
Argonne IL  60439                        |  as a last resort, to protect
                                         |  themselves against tyranny in
woyna@anl.gov                            |  government" -- Thomas Jefferson
                                         |
===============================================================================



