I am hitting a case where the blt smoke test passes, but I believe the MPI environment is actually bad.

I suggest we enhance the blt smoke test to check the size of the communicator to make sure it matches the expected number of MPI ranks (which I believe is 4)

The MPI reduce in blt's mpi smoke does not fail, b/c I expect it is self consistent even if there is only one MPI rank.

My more complex (non blt) tests that use mpi send / recv fail with a message like:

Invalid rank has value 1 but must be nonnegative and less than 1

(this means the only valid rank id is 0)

These tests should be definitely be launched with more than 1 mpi task :-(.

It is also possible that my MPI tests are launched differently than the BLT MPI test and that is where the surprise is.

Either way, having the comm size check in blt mpi smoke would provide more info and confidence about the underlying issue.