Mentions légales du service

Skip to content

connect_to_launcher() MPI Barrier issues when reader ranks > 2

PURANDARE Abhishek requested to merge server-rank-issues into adios2-newapi

Bug: MPI Barrier() when connecting to the launcher while initializing connections. This came after I have been using guix profile. May require some openmpi updates.

2024-05-24 16:39:49,156:melissa.server.main:ERROR Server failed with msg MPI_ERR_OTHER: known error not in list.
Traceback (most recent call last):
  File "/home/apurandare/MELISSA/melissa/server/main.py", line 100, in main
    myserver.initialize_connections()
  File "/home/apurandare/MELISSA/melissa/server/base_server.py", line 196, in initialize_connections
    self.connect_to_launcher()
  File "/home/apurandare/MELISSA/melissa/server/base_server.py", line 236, in connect_to_launcher
    self.comm.Barrier()
  File "mpi4py/MPI/Comm.pyx", line 675, in mpi4py.MPI.Comm.Barrier
mpi4py.MPI.Exception: MPI_ERR_OTHER: known error not in list

This issue is coming from OpenMPI 4.1.6 but only on the Infiniband node.

Edited by PURANDARE Abhishek

Merge request reports