Charm++ parallel programming system

[Matthias Diener <mdiener AT illinois.edu>]

Hi Jozsef,

We recently implemented a way to pass optional exit codes to CkExit.
Using the current git version of charm++, you can use CkExit(int
exitcode) to pass a nonzero exit code during termination.

Regards,
Matthias

On Thu, 17 May 2018 at 16:13, Matthias Diener
<mdiener AT illinois.edu>
 wrote:
>
> Hi Jozsef,
>
> Unfortunately, I think there is currently no way to pass a custom exit code
> to CkExit() or CkAbort().
> We have an open bug report for this (
> https://charm.cs.illinois.edu/redmine/issues/1584).
>
> As a temporary workaround, you could try calling abort() on an error
> condition.
>
> Regards,
> Matthias
>
> On Thu, 17 May 2018 at 16:07, Jozsef Bakosi 
> <jbakosi AT lanl.gov>
>  wrote:
>
> > Thanks, Matthias, and Nils, that works.
>
> > If I don't call CkAbort(), is there any other way to terminate a Charm++
> program
> > so I can get a nonzero exit code? CkAbort returns 1, which I would like
> to avoid
> > because I'm generating my own backtrace and CkExit() returns 0. Ideally,
> e.g., I
> > would like to pass a nonzero exit code CkExit().
>
> > Jozsef
>
> > On 05.16.2018 12:10, Matthias Diener wrote:
> > > Hi Jozsef,
> > >
> > > A bit of context regarding the second part of your question:
> > > Since https://charm.cs.illinois.edu/gerrit/#/c/charm/+/3720/ , Charm++
> > > calls CkAbort() when there are unhandled C++ exceptions in user code.
> > > Charm++ itself does not use exceptions.
> > >
> > > You could override Charm++'s uncaught exception handler in the
> mainchare,
> > > for
> > > example, by specifying your own set_terminate handler:
> > >
> > > // in Main():
> > > std::set_terminate([](){ CkPrintf("Warning: Unhandled exception in user
> > > code.\n"); abort();});
> > >
> > > Regards,
> > > Matthias
> > > On Wed, 16 May 2018 at 08:09, Jozsef Bakosi 
> > > <jbakosi AT lanl.gov>
> > >  wrote:
> > >
> > > > Hi folks,
> > >
> > > > When exceptions are thrown from chare arrays or groups, how/where one
> is
> > > > supposed to catch them?
> > >
> > > > Ideally, I would like to catch exceptions at the outermost layer,
> e.g.,
> > > outside
> > > > of some driver, from where everything else is instantiated and called,
> > > but that
> > > > way not every exception is caught.
> > >
> > > > I attach an example with the simplearrayhello test augmented and
> commented
> > > > highlighting the problem/question.
> > >
> > > > In addition, I also set signal handlers to throw exceptions (and
> produce
> > > nice
> > > > stack traces) when, e.g., segfaults happen and redirect them to the
> same
> > > code
> > > > that handles exceptions and produce stack traces, but when exceptions
> are
> > > not
> > > > caught that is useless.
> > >
> > > > Alternatively, is it possible to override Charm++'s exception
> handling,
> > > that
> > > > sometimes produces outputs like the following:
> > >
> > > > ------------- Processor 0 Exiting: Called CmiAbort ------------
> > > > Reason: Unhandled C++ exception in user code.
> > >
> > > > [0] Stack Traceback:
> > > >     [0:0]   [0x592cfe]
> > > >     [0:1]   [0x58f9f8]
> > > >     [0:2]   [0x41b908]
> > > >     [0:3]   [0x41b8e9]
> > > >     [0:4] +0x22763  [0x7ffff736d763]
> > > >     [0:5] +0x25516  [0x7ffff7370516]
> > > >     [0:6] +0x254af  [0x7ffff73704af]
> > > >     [0:7]   [0x414829]
> > > >     [0:8]   [0x413091]
> > > >     [0:9]   [0x434f30]
> > > >     [0:10]   [0x46f99a]
> > > >     [0:11]   [0x440c2d]
> > > >     [0:12]   [0x437491]
> > > >     [0:13]   [0x436e11]
> > > >     [0:14]   [0x596417]
> > > >     [0:15]   [0x596744]
> > > >     [0:16]   [0x59646a]
> > > >     [0:17]   [0x5929d1]
> > > >     [0:18]   [0x592283]
> > > >     [0:19]   [0x417c95]
> > > >     [0:20] __libc_start_main+0xe7  [0x7ffff621aa87]
> > > >     [0:21]   [0x411f1a]
> > > >
> --------------------------------------------------------------------------
> > > > MPI_ABORT was invoked on rank 0 in communicator MPI COMMUNICATOR 3 DUP
> > > FROM 0
> > > > with errorcode 1.
> > >
> > > > NOTE: invoking MPI_ABORT causes Open MPI to kill all MPI processes.
> > > > You may or may not see output from other processes, depending on
> > > > exactly when Open MPI kills them.
> > > >
> --------------------------------------------------------------------------
> > >
> > > > Thanks,
> > > > Jozsef