[MPICH] Crash While Spawning Process

[David Ashton]

Jason,

After modifying your example to correct syntax errors I ran the job like
this:

C:\Temp>"\Program Files\MPICH2\bin\mpiexec.exe" -exitcodes -n 1
\\grass\c$\temp\myapp.exe
Parent com: 4000000
Error: This MPI Implementation doesn't support universe size.
Spawning...
\\grass\c$\temp\myapp.exe
Parent com: 84000000
Spawned.
rank: node: exit code
0: hopper: 0
Parent com: 84000001
Enter the number of intervals: (0 quits) 342
pi is approximately 3.1415933660597268, Error is 0.0000007124699337
wall clock time = 0.002582
Enter the number of intervals: (0 quits) 0
rank: node: exit code
0: hopper: 0
1: hopper: 0

Everything went fine without any crashing.

But you aren't going to get a valid value for universe size.  MPICH2 for
Windows currently doesn't support that value.

-David Ashton

Here's the code I ran:

/* -*- Mode: C; c-basic-offset:4 ; -*- */
/*
 *  (C) 2001 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

/* This is an interactive version of cpi */
#include "mpi.h"
#include <stdio.h>
#include <math.h>

double f(double);

double f(double a)
{
    return (4.0 / (1.0 + a*a));
}

int main(int argc,char *argv[])
{
    int done = 0, n, myid, numprocs, i;
    double PI25DT = 3.141592653589793238462643;
    double mypi, pi, h, sum, x;
    double startwtime = 0.0, endwtime;
    int  namelen;
    char processor_name[MPI_MAX_PROCESSOR_NAME];
    MPI_Comm parentComm;

    MPI_Init(&argc,&argv);
    MPI_Comm_size(MPI_COMM_WORLD,&numprocs);
    MPI_Comm_rank(MPI_COMM_WORLD,&myid);
    MPI_Get_processor_name(processor_name,&namelen);

    MPI_Comm_get_parent(&parentComm);
    printf("Parent com: %X\n", parentComm); fflush(stdout);

    if (myid == 0 && parentComm == MPI_COMM_NULL)
    {
        int universeSize, *universeSizep;
        int flag;
        int result;
        MPI_Comm everyone;

        MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universeSizep,
&flag);
        if (!flag)
        {
            printf("Error: This MPI Implementation doesn't support universe
size.\n"); fflush(stdout);
            universeSize = 2;
        }
        else
        {
            universeSize = *universeSizep;
        }

        printf("Spawning...\n"); fflush(stdout);

        printf("%s\n", argv[0]); fflush(stdout);
        result = MPI_Comm_spawn(argv[0], MPI_ARGV_NULL, universeSize,
            MPI_INFO_NULL, 0, MPI_COMM_SELF, &everyone,
MPI_ERRCODES_IGNORE); 

        printf("Spawned.\n"); fflush(stdout);

        MPI_Finalize();
        return 0;
    }

    /*
    fprintf(stdout,"Process %d of %d is on %s\n",
	    myid, numprocs, processor_name);
    fflush(stdout);
    */

    while (!done) {
        if (myid == 0) {
            fprintf(stdout, "Enter the number of intervals: (0 quits) ");
	    fflush(stdout);
            if (scanf("%d",&n) != 1) {
		fprintf( stdout, "No number entered; quitting\n" );
		n = 0;
	    }
	    startwtime = MPI_Wtime();
        }
        MPI_Bcast(&n, 1, MPI_INT, 0, MPI_COMM_WORLD);
        if (n == 0)
            done = 1;
        else {
            h   = 1.0 / (double) n;
            sum = 0.0;
            for (i = myid + 1; i <= n; i += numprocs) {
                x = h * ((double)i - 0.5);
                sum += f(x);
            }
            mypi = h * sum;
            MPI_Reduce(&mypi, &pi, 1, MPI_DOUBLE, MPI_SUM, 0,
MPI_COMM_WORLD);

            if (myid == 0) {
                printf("pi is approximately %.16f, Error is %.16f\n",
                       pi, fabs(pi - PI25DT));
		endwtime = MPI_Wtime();
		printf("wall clock time = %f\n", endwtime-startwtime);

		fflush( stdout );
	    }
        }
    }
    MPI_Finalize();
    return 0;
}


-----Original Message-----
From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Jason Surratt
Sent: Monday, October 17, 2005 7:28 AM
To: MPICH Mailing List
Subject: [MPICH] Crash While Spawning Process


Hello,

I am new to MPI, but I am very happy with the initial success I've had 
parallelizing my application with MPICH2.

The problem I'm having is when I spawn a process with the MPI_Comm_spawn 
function. To keep things simple I have modified the icpi.c example to 
include the following code snippet after line 35:

...
  MPI_Comm parentComm;
  MPI_Comm_get_parent(&parentComm);
  printf("Parent com: %X\n", parentComm);

  if (myid == 0 && parentComm == MPI_COMM_NULL)
  {
    int universeSize, *universeSizep;
    int flag;
    MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universeSizep, 
&flag);
    if (!flag)
    {
      printf("Error: This MPI Implementation doesn't support universe 
size.\n"); fflush(stdout);
      universeSize = 2;
    }
    else
    {
      universeSize = *universeSizep;
    }

    int result;
    MPI_Comm everyone;

    printf("Spawning...\n"); fflush(stdout);

    printf("%s\n", argv[0]); fflush(stdout);
    result = MPI_Comm_spawn(argv[0], MPI_ARGV_NULL, universeSize,
      MPI_INFO_NULL, 0, MPI_COMM_SELF, &everyone, MPI_ERRCODES_IGNORE); 

    printf("Spawned.\n"); fflush(stdout);

    MPI_Finalize();
    return 0;
  }
...

I then launch this application with:
mpiexec -exitcodes -n 1 \\jsurratt-laptop\simple_mpi\simple_mpi

My intent is to launch universeSize processes without requiring the user 
to type the size at the command line.

If I run this code with my local machine as the only host the universe 
size is not returned (flag == 0), but the application spawns two 
processes and runs fine.

If I add another similarly configured machine to my host list the 
universe size is not returned (flag == 0) and I get the standard Windows 
crash prompt 'Process launcher for MPICH2 applications has encountered a 
problem and needs to close.  We are sorry for the inconvenience.' when 
the MPI_Comm_spawn function is called.

I'm hoping that I'm just doing something silly, but if need be I can 
bring it up in the debugger and post more information.

I'm using the following configuration on all machines:
Windows XP SP2
MPICH2 v1.0.2-1 (ia32)

Running the unmodified version of icpi.c with multiple hosts works fine.

Thanks in advance,

-Jason