Re: [OMPI users] MPI_Spawn error: Data unpack would read past end of buffer" (-26) instead of "Success"

Simone Pellegrini Wed, 7 Sep 2011 14:03:32 -0400

On 09/06/2011 06:11 PM, Ralph Castain wrote:

Hmmm...well, nothing definitive there, I'm afraid.


All I can suggest is to remove/reduce the threading. Like I said, we aren't 
terribly thread safe at this time. I suspect you're stepping into one of those 
non-safe areas here.

Hopefully will do better in later releases.

Hi again,

I made some improvements on this problem myself. It looks like is not arelated to threading and/or race conditions but instead to the behaviorof MPI_Finalize invoked by the spawned processes. Apparently despite thespawned processes all invoke MPI_Finalize, the processes remains aliveblocked on a semaphore. Therefore by spawning more and more processes Iend up having hundreds of processes and slowly filling up all theavailable file descriptors.

I got this hint by running my code with mpich2. After a while I also getan error there related to file descriptors and since then it was easy tounderstand what was going on (you should made errors semantically moresound in open mpi).

By the way, I solved the problem by invoking MPI_Comm_disconnect on theinter-communicator I receive from the spawning task (MPI_Finalize is notenough). This makes the spawned tasks to close the parent communicatorand terminate.

After this small change the system is more stable now and that specificerror is gone. Unfortunately a different message showed up:

[arch-moto][[530,1],0][btl_tcp_frag.c:215:mca_btl_tcp_frag_recv]mca_btl_tcp_frag_recv: readv failed: Bad file descriptor (9)


This error doesn't make the program terminate.

Some other times I get an hard error, which is:
[err] event_queue_remove: 0x7fb5fc008c58(fd 14) not on queue 8

[arch-moto][[14492,46],0][btl_tcp_endpoint.c:638:mca_btl_tcp_endpoint_complete_connect]connect() to 192.168.88.1 failed: Connection refused (111)[arch-moto:09536] [[14492,0],0] ORTE_ERROR_LOG: A message is attemptingto be sent to a process whose contact information is unknown in filerml_oob_send.c at line 145

[arch-moto:09536] [[14492,0],0] attempted to send to [[14492,1],0]: tag 6

[arch-moto:09536] [[14492,0],0] ORTE_ERROR_LOG: A message is attemptingto be sent to a process whose contact information is unknown in filebase/plm_base_receive.c at line 278

--------------------------------------------------------------------------
mpirun has exited due to process rank 0 with PID 9538 on
node arch-moto exiting improperly. There are two reasons this could occur:

1. this process did not call "init" before exiting, but others in
the job did. This can cause a job to hang indefinitely while it waits
for all processes to call "init". By rule, if one process calls "init",
then ALL processes must call "init" prior to termination.

2. this process called "init", but exited without calling "finalize".
By rule, all processes that call "init" MUST call "finalize" prior to
exiting or it will be considered an "abnormal termination"

This may have caused other processes in the application to be
terminated by signals sent by mpirun (as reported here).
--------------------------------------------------------------------------

any hints from this?

cheers, Simone


On Sep 6, 2011, at 1:20 PM, Simone Pellegrini wrote:

On 09/06/2011 04:58 PM, Ralph Castain wrote:

On Sep 6, 2011, at 12:49 PM, Simone Pellegrini wrote:

On 09/06/2011 02:57 PM, Ralph Castain wrote:

Hi Simone

Just to clarify: is your application threaded? Could you please send the OMPI 
configure cmd you used?

yes, it is threaded. There are basically 3 threads, 1 for the outgoing messages 
(MPI_send), 1 for incoming messages (MPI_Iprobe / MPI_Recv) and one spawning.

I am not sure what you mean with OMPI configure cmd I used... I simply do 
mpirun --np 1 ./executable

How was OMPI configured when it was installed? If you didn't install it, then 
provide the output of ompi_info - it will tell us.

[@arch-moto tasksys]$ ompi_info
                 Package: Open MPI nobody@alderaan Distribution
                Open MPI: 1.5.3
   Open MPI SVN revision: r24532
   Open MPI release date: Mar 16, 2011
                Open RTE: 1.5.3
   Open RTE SVN revision: r24532
   Open RTE release date: Mar 16, 2011
                    OPAL: 1.5.3
       OPAL SVN revision: r24532
       OPAL release date: Mar 16, 2011
            Ident string: 1.5.3
                  Prefix: /usr
Configured architecture: x86_64-unknown-linux-gnu
          Configure host: alderaan
           Configured by: nobody
           Configured on: Thu Jul  7 13:21:35 UTC 2011
          Configure host: alderaan
                Built by: nobody
                Built on: Thu Jul  7 13:27:08 UTC 2011
              Built host: alderaan
              C bindings: yes
            C++ bindings: yes
      Fortran77 bindings: yes (all)
      Fortran90 bindings: yes
Fortran90 bindings size: small
              C compiler: gcc
     C compiler absolute: /usr/bin/gcc
  C compiler family name: GNU
      C compiler version: 4.6.1
            C++ compiler: g++
   C++ compiler absolute: /usr/bin/g++
      Fortran77 compiler: gfortran
  Fortran77 compiler abs: /usr/bin/gfortran
      Fortran90 compiler: /usr/bin/gfortran
  Fortran90 compiler abs:
             C profiling: yes
           C++ profiling: yes
     Fortran77 profiling: yes
     Fortran90 profiling: yes
          C++ exceptions: no
          Thread support: posix (mpi: yes, progress: no)
           Sparse Groups: no
  Internal debug support: yes
  MPI interface warnings: no
     MPI parameter check: runtime
Memory profiling support: no
Memory debugging support: no
         libltdl support: yes
   Heterogeneous support: no
mpirun default --prefix: no
         MPI I/O support: yes
       MPI_WTIME support: gettimeofday
     Symbol vis. support: yes
          MPI extensions: affinity example
   FT Checkpoint support: no (checkpoint thread: no)
  MPI_MAX_PROCESSOR_NAME: 256
    MPI_MAX_ERROR_STRING: 256
     MPI_MAX_OBJECT_NAME: 64
        MPI_MAX_INFO_KEY: 36
        MPI_MAX_INFO_VAL: 256
       MPI_MAX_PORT_NAME: 1024
  MPI_MAX_DATAREP_STRING: 128
           MCA backtrace: execinfo (MCA v2.0, API v2.0, Component v1.5.3)
          MCA memchecker: valgrind (MCA v2.0, API v2.0, Component v1.5.3)
              MCA memory: linux (MCA v2.0, API v2.0, Component v1.5.3)
           MCA paffinity: hwloc (MCA v2.0, API v2.0, Component v1.5.3)
               MCA carto: auto_detect (MCA v2.0, API v2.0, Component v1.5.3)
               MCA carto: file (MCA v2.0, API v2.0, Component v1.5.3)
           MCA maffinity: first_use (MCA v2.0, API v2.0, Component v1.5.3)
               MCA timer: linux (MCA v2.0, API v2.0, Component v1.5.3)
         MCA installdirs: env (MCA v2.0, API v2.0, Component v1.5.3)
         MCA installdirs: config (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA dpm: orte (MCA v2.0, API v2.0, Component v1.5.3)
              MCA pubsub: orte (MCA v2.0, API v2.0, Component v1.5.3)
           MCA allocator: basic (MCA v2.0, API v2.0, Component v1.5.3)
           MCA allocator: bucket (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: basic (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: hierarch (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: inter (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: self (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: sm (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: sync (MCA v2.0, API v2.0, Component v1.5.3)
                MCA coll: tuned (MCA v2.0, API v2.0, Component v1.5.3)
                  MCA io: romio (MCA v2.0, API v2.0, Component v1.5.3)
               MCA mpool: fake (MCA v2.0, API v2.0, Component v1.5.3)
               MCA mpool: rdma (MCA v2.0, API v2.0, Component v1.5.3)
               MCA mpool: sm (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA pml: bfo (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA pml: csum (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA pml: ob1 (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA pml: v (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA bml: r2 (MCA v2.0, API v2.0, Component v1.5.3)
              MCA rcache: vma (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA btl: self (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA btl: sm (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA btl: tcp (MCA v2.0, API v2.0, Component v1.5.3)
                MCA topo: unity (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA osc: pt2pt (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA osc: rdma (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA iof: hnp (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA iof: orted (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA iof: tool (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA oob: tcp (MCA v2.0, API v2.0, Component v1.5.3)
                MCA odls: default (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA ras: cm (MCA v2.0, API v2.0, Component v1.5.3)
               MCA rmaps: load_balance (MCA v2.0, API v2.0, Component v1.5.3)
               MCA rmaps: rank_file (MCA v2.0, API v2.0, Component v1.5.3)
               MCA rmaps: resilient (MCA v2.0, API v2.0, Component v1.5.3)
               MCA rmaps: round_robin (MCA v2.0, API v2.0, Component v1.5.3)
               MCA rmaps: seq (MCA v2.0, API v2.0, Component v1.5.3)
               MCA rmaps: topo (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA rml: oob (MCA v2.0, API v2.0, Component v1.5.3)
              MCA routed: binomial (MCA v2.0, API v2.0, Component v1.5.3)
              MCA routed: cm (MCA v2.0, API v2.0, Component v1.5.3)
              MCA routed: direct (MCA v2.0, API v2.0, Component v1.5.3)
              MCA routed: linear (MCA v2.0, API v2.0, Component v1.5.3)
              MCA routed: radix (MCA v2.0, API v2.0, Component v1.5.3)
              MCA routed: slave (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA plm: rsh (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA plm: rshd (MCA v2.0, API v2.0, Component v1.5.3)
               MCA filem: rsh (MCA v2.0, API v2.0, Component v1.5.3)
              MCA errmgr: default (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA ess: env (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA ess: hnp (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA ess: singleton (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA ess: slave (MCA v2.0, API v2.0, Component v1.5.3)
                 MCA ess: tool (MCA v2.0, API v2.0, Component v1.5.3)
             MCA grpcomm: bad (MCA v2.0, API v2.0, Component v1.5.3)
             MCA grpcomm: basic (MCA v2.0, API v2.0, Component v1.5.3)
             MCA grpcomm: hier (MCA v2.0, API v2.0, Component v1.5.3)
            MCA notifier: command (MCA v2.0, API v1.0, Component v1.5.3)
            MCA notifier: syslog (MCA v2.0, API v1.0, Component v1.5.3)

Adding the debug flags just changes the race condition. Interestingly, those 
values only impact the behavior of mpirun, so it looks like the race condition 
is occurring there.

The problem is that the error is totally nondeterministic. Sometimes happens, 
others not but the error message gives me no clue where the error is coming 
from. Is is a problem of my code or internal MPI?

Can't tell, but it is likely an impact of threading. Race conditions within 
threaded environments are common, and OMPI isn't particularly thread safe, 
especially when it comes to comm_spawn.

cheers, Simone

On Sep 6, 2011, at 3:01 AM, Simone Pellegrini wrote:

Dear all,
I am developing an MPI application which uses heavily MPI_Spawn. Usually 
everything works fine for the first hundred spawn but after a while the 
application exist with a curious message:

[arch-top:27712] [[36904,165],0] ORTE_ERROR_LOG: Data unpack would read past 
end of buffer in file base/grpcomm_base_modex.c at line 349
[arch-top:27712] [[36904,165],0] ORTE_ERROR_LOG: Data unpack would read past 
end of buffer in file grpcomm_bad_module.c at line 518
--------------------------------------------------------------------------
It looks like MPI_INIT failed for some reason; your parallel process is
likely to abort.  There are many reasons that a parallel process can
fail during MPI_INIT; some of which are due to configuration or environment
problems.  This failure appears to be an internal failure; here's some
additional information (which may only be relevant to an Open MPI
developer):

  ompi_proc_set_arch failed
  -->    Returned "Data unpack would read past end of buffer" (-26) instead of 
"Success" (0)
--------------------------------------------------------------------------
*** The MPI_Init_thread() function was called before MPI_INIT was invoked.
*** This is disallowed by the MPI standard.
*** Your MPI job will now abort.
[arch-top:27712] Abort before MPI_INIT completed successfully; not able to 
guarantee that all other processes were killed!
[arch-top:27714] [[36904,165],0] ORTE_ERROR_LOG: Data unpack would read past 
end of buffer in file base/grpcomm_base_modex.c at line 349
[arch-top:27714] [[36904,165],0] ORTE_ERROR_LOG: Data unpack would read past 
end of buffer in file grpcomm_bad_module.c at line 518
*** The MPI_Init_thread() function was called before MPI_INIT was invoked.
*** This is disallowed by the MPI standard.
*** Your MPI job will now abort.
[arch-top:27714] Abort before MPI_INIT completed successfully; not able to 
guarantee that all other processes were killed!
[arch-top:27226] 1 more process has sent help message help-mpi-runtime / 
mpi_init:startup:internal-failure
[arch-top:27226] Set MCA parameter "orte_base_help_aggregate" to 0 to see all 
help / error messages

Also using MPI_init instead of MPI_Init_thread does not help, the same error 
occurs.

Strangely the error does not occur if I run the code enabling debug in (-mca 
plm_base_verbose 5 -mca rmaps_base_verbose 5).

I am using OpenMPI 1.5.3

cheers, Simone
_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users


_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

Re: [OMPI users] MPI_Spawn error: Data unpack would read past end of buffer" (-26) instead of "Success"

Reply via email to