So as other's have mentioned on this mailing list before, one "gotcha" with
upgrading from 14.03.x to 14.11.x is that libpmi.so is linked to
libslurm.so.27 in 14.03.x.  In 14.11 the libslurm is named to
libslurm.so.28.  This can cause issues if OpenMPI or MVAPICH2 were compiled
with SLURM support.  I verified this is the case while testing the upgrade
to 14.11.6 at my site.  What I could use some input on is how safe is it to
just add the libslurm.so.27 symlink to the RPM spec used to build SLURM?  I
manually added the symlink on my test compute nodes then ran an MPI job
using OpenMPI and MVAPICH2 both compiled with SLURM 14.03 support.  With
the symlink manually created, the jobs ran fine under 14.11.6.  I'm curious
if there may be some hidden problem I'd hit later if I don't recompile all
my installs of OpenMPI and MVAPICH2.

Thanks,
- Trey

=============================

Trey Dockendorf
Systems Analyst I
Texas A&M University
Academy for Advanced Telecommunications and Learning Technologies
Phone: (979)458-2396
Email: [email protected]
Jabber: [email protected]

Reply via email to