diff --git a/m/common/slurm.nix b/m/common/slurm.nix index 650156f..ce909df 100644 --- a/m/common/slurm.nix +++ b/m/common/slurm.nix @@ -10,7 +10,15 @@ "hut Sockets=2 CoresPerSocket=14 ThreadsPerCore=2" ]; extraConfig = '' + # Use PMIx for MPI by default. It works okay with MPICH and OpenMPI, but + # not with Intel MPI. For that use the compatibility shim libpmi.so + # setting I_MPI_PMI_LIBRARY=$pmix/lib/libpmi.so while maintaining the PMIx + # library in SLURM (--mpi=pmix). See more details here: + # https://pm.bsc.es/gitlab/rarias/jungle/-/issues/16 MpiDefault=pmix + + # When a node reboots return that node to the slurm queue as soon as it + # becomes operative again. ReturnToService=2 ''; };