[MPICH2-dev] spawn and MPI_Finalize

[Rajeev Thakur]

Are you using the latest version, 1.0.6p1? At least the issue of the slaves
waiting for the parent to call finalize should be fixed in there.

Rajeev

> -----Original Message-----
> From: owner-mpich2-dev at mcs.anl.gov 
> [mailto:owner-mpich2-dev at mcs.anl.gov] On Behalf Of 
> Rozzen.VINCONT at external.thalesgroup.com
> Sent: Thursday, November 22, 2007 1:31 PM
> To: mpich2-dev at mcs.anl.gov
> Subject: [MPICH2-dev] Date: Thu, 22 Nov 2007 09:51:08 +0100
> 
> Hi,
> I have some issues using MPI_Finalize.
> 
> Issue 1:
> --------
> A master process spawns 2 slave processes. Slave processes 
> perform a job and then stop.
> 
> My problem is one of the slave process is stuck in its 
> MPI_Finalize call until Master reaches its own MPI_Finalize.
> 
> Why a slave is stuck in its MPI_Finalize call ? why the other is not ?
> 
> Note: by the way, with 10 slaves spawned from a single 
> MPI_spawn_multiple call, only 1 out of 10 is stuck.
> 
> 
> Issue 2:
> --------
> A master process spawns every second a slave process (Master 
> does not know if Slave is stopped or not before doing a spawn again).
> 
> Conf. 1: Slave is stuck in its MPI_Finalize call.
> In this configuration, Master is able to perform 324 spawn 
> calls and then Master stops with no obvious reason.
> 
> Conf. 2: MPI_Finalize is removed from the slave source code. 
> Slave stops correctly but Master reports a communication shutdown.
> In this configuration, Master is able to perform about 550 
> spawn calls and then Master stops with no obvious reason.
> 
> Do you have any explanations for these spawn limits ?
> 
> 
> Thanks for your help.
> 
> Regards.
> 
>