[mpich-discuss] _get_addr error while running application using MPICH

Balaji, Pavan balaji at anl.gov
Wed Nov 21 17:34:55 CST 2018


I should clarify one piece of how mpiexec works:

% ./real.exe

and

% mpirun -np 1 ./real.exe

are equivalent.

% mpirun ./real.exe

uses some number of processes depending on the environment.  For unmanaged clusters, that's typically 1.  For clusters that have some job management system (such as slurm or pbs), mpiexec will figure out how many nodes you allocated and use all of the cores allocated to that job.

My guess is that real.exe has some dependencies that are met on the local machine, but not on other machines.  So when mpiexec tries to launch real.exe on other nodes, it's throwing an error.  This is not an mpiexec problem, but you might want to use the -prepend-pattern option in mpiexec to figure out where the error is coming from.  Something like this:

mpiexec -prepend-pattern %h ./real.exe

  -- Pavan

> On Nov 4, 2018, at 10:27 AM, Zhifeng Yang via discuss <discuss at mpich.org> wrote:
> 
> Hi
> 
> After I installed mpich and used it in a FORTRAN code. There is an error while running this FORTRAN code called real.exe by using the following command
> 
> $./real.exe
> or
> $mpirun ./real.exe
> real.exe: error: _get_addr: No such file or directory
> real.exe: error: _get_addr: No such file or directory
> real.exe: error: _get_addr: No such file or directory
> Fatal error in MPI_Init: Other MPI error, error stack:
> MPIR_Init_thread(784).....:
> MPID_Init(1323)...........: channel initialization failed
> MPIDI_CH3_Init(120).......:
> MPID_nem_init_ckpt(852)...:
> MPIDI_CH3I_Seg_commit(364): PMI_Barrier returned -1
>  
> But while using mpirun with specifying number of processors. as follows
> $mpirun -np 1 ./real.exe
> There is no error any more. I am not sure why? do you have any explanation? Thank you very much
> 
> Best regards
> Zhifeng
> 
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss




More information about the discuss mailing list