It looks like a garden variety race condition -- MCW rank 1 is probably calling
MPI_Iprobe before the message has actually been delivered by MCW rank 0.
Try putting in a sleep(1) (or some other kind of delay) before your MPI_Iprobe,
and I'll bet that the first call to MPI_Iprobe will return has_message=1.
Or just use MPI_Probe -- it blocks, unlike MPI_Iprobe.
On Jul 22, 2011, at 4:29 PM, Rodrigo Oliveira wrote:
> Hi there.
>
> I have an application in which I need to terminate a process anytime due an
> external command. In order to maintain the consistence of the processes, I
> need to receive the messages that were already sent to the terminating
> process. I used the MPI_Iprobe to check whether there is messages to be
> received, but I noticed that I have to call this function twice. Otherwise it
> does not work properly. The code bellow exemplifies what happens. Can anyone
> help me? Is there another way to do what I need?
>
> Thanks in advance.
>
>
> #include "mpi.h"
> #include <stdio.h>
>
> int main(int argc, char *argv[]) {
> int rank, size, i;
> MPI_Status status;
>
> MPI_Init(&argc, &argv);
> MPI_Comm_size(MPI_COMM_WORLD, &size);
> MPI_Comm_rank(MPI_COMM_WORLD, &rank);
>
> if (size < 2) {
> printf("Please run with two processes.\n"); fflush(stdout);
> MPI_Finalize();
> return 0;
> }
> if (rank == 0) {
> for (i=0; i<10; i++) {
> MPI_Send(&i, 1, MPI_INT, 1, 123, MPI_COMM_WORLD);
> }
> }
> if (rank == 1) {
> int value, has_message;
> MPI_Status status;
> sleep (2);
>
> // Code bellow does not work properly
> MPI_Iprobe(0, 123, MPI_COMM_WORLD, &has_message, &status);
> while (has_message) {
> MPI_Recv(&value, 1, MPI_INT, 0, 123, MPI_COMM_WORLD,
> &status);
> printf("Process %d received message %d.\n", rank,
> value);
> MPI_Iprobe(0, 123, MPI_COMM_WORLD, &has_message,
> &status);
> }
>
> // Calling MPI_Iprobe twice for each incoming message makes the
> code work.
> /*
> MPI_Iprobe(0, 123, MPI_COMM_WORLD, &has_message, &status);
> MPI_Iprobe(0, 123, MPI_COMM_WORLD, &has_message, &status);
> while (has_message) {
> MPI_Recv(&value, 1, MPI_INT, 0, 123, MPI_COMM_WORLD,
> &status);
> printf("Process %d received message %d.\n", rank,
> value);
> MPI_Iprobe(0, 123, MPI_COMM_WORLD, &has_message,
> &status);
> MPI_Iprobe(0, 123, MPI_COMM_WORLD, &has_message,
> &status);
> }
> */
>
> fflush(stdout);
> }
> MPI_Finalize();
> return 0;
> }
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users
--
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/
