[mpich-discuss] Catching Exceptions

[Darius Buntinas]

Currently, mpich2 does not recover from a process failure, so the entire
job is killed, even when ERRORS_THROW_EXCEPTION is set.  However, we are
actively working on this issue.

Darius

On 08/18/2009 05:38 PM, MSR wrote:
> Hi,
> 
> I noticed that when a process dies then MPIException is not thrown on
> other process. They all abort with message "rank N caused collective
> abort of all ranks". Is this how it is supposed to work? Or should I
> doing to something special so that MPI throws and exception. I even
> tried with MPI::COMM_WORLD.Set_errhandler(MPI::ERRORS_THROW_EXCEPTION)
> but still no exceptions raised in the case where a process dies.
> 
> Thanks,
> MSR