[petsc-users] Reaching limit number of communicator with Spectrum MPI

Feimi Yu yuf2 at rpi.edu
Wed Aug 18 16:23:52 CDT 2021


Hi Satish and Junchao,

I just tried replacing all MPI_COMM_WORLD with PETSC_COMM_WORLD, but it 
didn't do the trick. One thing that interests me is that, I ran with 40 
ranks but only 2 ranks reported the communicator error. I think this 
means at least the rest 38 ranks freed the communicators properly.

Thanks!

Feimi

On 8/18/21 4:53 PM, Junchao Zhang wrote:
> Hi, Feimi,
>   I need to consult Jed (cc'ed).
>   Jed, is this an example of 
> https://lists.mcs.anl.gov/mailman/htdig/petsc-dev/2018-April/thread.html#22663 
> <https://lists.mcs.anl.gov/mailman/htdig/petsc-dev/2018-April/thread.html#22663>? 
> If Feimi really can not free matrices, then we just need to attach a 
> hypre-comm to a petsc inner comm, and pass that to hypre.
>
> --Junchao Zhang
>
>
> On Wed, Aug 18, 2021 at 3:38 PM Satish Balay <balay at mcs.anl.gov 
> <mailto:balay at mcs.anl.gov>> wrote:
>
>     Is the communicator used to create PETSc objects MPI_COMM_WORLD?
>
>     If so - try changing it to PETSC_COMM_WORLD
>
>     Satish
>
>      On Wed, 18 Aug 2021, Feimi Yu wrote:
>
>     > Hi Junchao,
>     >
>     > Thank you for the suggestion! I'm using the deal.ii wrapper
>     > dealii::PETScWrappers::PreconditionBase to handle the PETSc
>     preconditioners,
>     > and the wrappers does the destroy when the preconditioner is
>     reinitialized or
>     > gets out of scope. I just double-checked, this is called to make
>     sure the old
>     > matrices are destroyed:
>     >
>     >    void
>     >    PreconditionBase::clear()
>     >    {
>     >      matrix = nullptr;
>     >
>     >      if (pc != nullptr)
>     >        {
>     >          PetscErrorCode ierr = PCDestroy(&pc);
>     >          pc                  = nullptr;
>     >          AssertThrow(ierr == 0, ExcPETScError(ierr));
>     >        }
>     >    }
>     >
>     > Thanks!
>     >
>     > Feimi
>     >
>     > On 8/18/21 4:23 PM, Junchao Zhang wrote:
>     > >
>     > >
>     > >
>     > > On Wed, Aug 18, 2021 at 12:52 PM Feimi Yu <yuf2 at rpi.edu
>     <mailto:yuf2 at rpi.edu>
>     > > <mailto:yuf2 at rpi.edu <mailto:yuf2 at rpi.edu>>> wrote:
>     > >
>     > >     Hi,
>     > >
>     > >     I was trying to run a simulation with a PETSc-wrapped Hypre
>     > >     preconditioner, and encountered this problem:
>     > >
>     > >     [dcs122:133012] Out of resources: all 4095 communicator
>     IDs have
>     > >     been used.
>     > >     [19]PETSC ERROR: --------------------- Error Message
>     > >  --------------------------------------------------------------
>     > >     [19]PETSC ERROR: General MPI error
>     > >     [19]PETSC ERROR: MPI error 17 MPI_ERR_INTERN: internal error
>     > >     [19]PETSC ERROR: See
>     > > https://www.mcs.anl.gov/petsc/documentation/faq.html
>     <https://www.mcs.anl.gov/petsc/documentation/faq.html>
>     > >     <https://www.mcs.anl.gov/petsc/documentation/faq.html
>     <https://www.mcs.anl.gov/petsc/documentation/faq.html>> for trouble
>     > >     shooting.
>     > >     [19]PETSC ERROR: Petsc Release Version 3.15.2, unknown
>     > >     [19]PETSC ERROR: ./main on a arch-linux-c-opt named dcs122 by
>     > >     CFSIfmyu Wed Aug 11 19:51:47 2021
>     > >     [19]PETSC ERROR: [dcs122:133010] Out of resources: all 4095
>     > >     communicator IDs have been used.
>     > >     [18]PETSC ERROR: --------------------- Error Message
>     > >  --------------------------------------------------------------
>     > >     [18]PETSC ERROR: General MPI error
>     > >     [18]PETSC ERROR: MPI error 17 MPI_ERR_INTERN: internal error
>     > >     [18]PETSC ERROR: See
>     > > https://www.mcs.anl.gov/petsc/documentation/faq.html
>     <https://www.mcs.anl.gov/petsc/documentation/faq.html>
>     > >     <https://www.mcs.anl.gov/petsc/documentation/faq.html
>     <https://www.mcs.anl.gov/petsc/documentation/faq.html>> for trouble
>     > >     shooting.
>     > >     [18]PETSC ERROR: Petsc Release Version 3.15.2, unknown
>     > >     [18]PETSC ERROR: ./main on a arch-linux-c-opt named dcs122 by
>     > >     CFSIfmyu Wed Aug 11 19:51:47 2021
>     > >     [18]PETSC ERROR: Configure options --download-scalapack
>     > >     --download-mumps --download-hypre --with-cc=mpicc
>     > >     --with-cxx=mpicxx --with-fc=mpif90 --with-cudac=0
>     > >     --with-debugging=0
>     > >
>      --with-blaslapack-dir=/gpfs/u/home/CFSI/CFSIfmyu/barn-shared/dcs-rh8/lapack-build/
>     > >     [18]PETSC ERROR: #1 <https://itssc.rpi.edu/hc/requests/1
>     <https://itssc.rpi.edu/hc/requests/1>>
>     > >     MatCreate_HYPRE() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/impls/hypre/mhypre.c:2120
>     > >     [18]PETSC ERROR: #2 <https://itssc.rpi.edu/hc/requests/2
>     <https://itssc.rpi.edu/hc/requests/2>>
>     > >     MatSetType() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/interface/matreg.c:91
>     > >     [18]PETSC ERROR: #3 <https://itssc.rpi.edu/hc/requests/3
>     <https://itssc.rpi.edu/hc/requests/3>>
>     > >     MatConvert_AIJ_HYPRE() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/impls/hypre/mhypre.c:392
>     > >     [18]PETSC ERROR: #4 <https://itssc.rpi.edu/hc/requests/4
>     <https://itssc.rpi.edu/hc/requests/4>>
>     > >     MatConvert() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/interface/matrix.c:4439
>     > >     [18]PETSC ERROR: #5 <https://itssc.rpi.edu/hc/requests/5
>     <https://itssc.rpi.edu/hc/requests/5>>
>     > >     PCSetUp_HYPRE() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/ksp/pc/impls/hypre/hypre.c:240
>     > >     [18]PETSC ERROR: #6 <https://itssc.rpi.edu/hc/requests/6
>     <https://itssc.rpi.edu/hc/requests/6>>
>     > >     PCSetUp() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/ksp/pc/interface/precon.c:1015
>     > >     Configure options --download-scalapack --download-mumps
>     > >     --download-hypre --with-cc=mpicc --with-cxx=mpicxx
>     > >     --with-fc=mpif90 --with-cudac=0 --with-debugging=0
>     > >
>      --with-blaslapack-dir=/gpfs/u/home/CFSI/CFSIfmyu/barn-shared/dcs-rh8/lapack-build/
>     > >     [19]PETSC ERROR: #1 <https://itssc.rpi.edu/hc/requests/1
>     <https://itssc.rpi.edu/hc/requests/1>>
>     > >     MatCreate_HYPRE() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/impls/hypre/mhypre.c:2120
>     > >     [19]PETSC ERROR: #2 <https://itssc.rpi.edu/hc/requests/2
>     <https://itssc.rpi.edu/hc/requests/2>>
>     > >     MatSetType() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/interface/matreg.c:91
>     > >     [19]PETSC ERROR: #3 <https://itssc.rpi.edu/hc/requests/3
>     <https://itssc.rpi.edu/hc/requests/3>>
>     > >     MatConvert_AIJ_HYPRE() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/impls/hypre/mhypre.c:392
>     > >     [19]PETSC ERROR: #4 <https://itssc.rpi.edu/hc/requests/4
>     <https://itssc.rpi.edu/hc/requests/4>>
>     > >     MatConvert() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/mat/interface/matrix.c:4439
>     > >     [19]PETSC ERROR: #5 <https://itssc.rpi.edu/hc/requests/5
>     <https://itssc.rpi.edu/hc/requests/5>>
>     > >     PCSetUp_HYPRE() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/ksp/pc/impls/hypre/hypre.c:240
>     > >     [19]PETSC ERROR: #6 <https://itssc.rpi.edu/hc/requests/6
>     <https://itssc.rpi.edu/hc/requests/6>>
>     > >     PCSetUp() at
>     > >
>      /gpfs/u/barn/CFSI/shared/dcs-rh8/petsc/src/ksp/pc/interface/precon.c:1015
>     > >
>     > >     It seems that MPI_Comm_dup() at
>     > >     petsc/src/mat/impls/hypre/mhypre.c:2120 caused the
>     problem. Since
>     > >     mine is a time-dependent problem, MatCreate_HYPRE() is called
>     > >     every time the new system matrix is assembled. The above error
>     > >     message is reported after ~4095 calls of
>     MatCreate_HYPRE(), which
>     > >     is around 455 time steps in my code. Here is some basic
>     compiler
>     > >     information:
>     > >
>     > > Can you destroy old matrices to free MPI communicators?
>     Otherwise, you run
>     > > into a limitation we knew before.
>     > >
>     > >     IBM Spectrum MPI 10.4.0
>     > >
>     > >     GCC 8.4.1
>     > >
>     > >     I've never had this problem before with OpenMPI or MPICH
>     > >     implementation, so I was wondering if this can be resolved
>     from my
>     > >     end, or it's an implementation specific problem.
>     > >
>     > >     Thanks!
>     > >
>     > >     Feimi
>     > >
>     >
>     > 
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20210818/18fccc09/attachment.html>


More information about the petsc-users mailing list