Documentation says yogrt_remaining can only be called on rank 0

[tepperly]

The documentation says that yogrt can only be called on process 0. Presumably process means rank in an MPI context?

libyogrt/doc/man/yogrt_remaining.3

Line 23 in 4346565

(This call may ONLY be used by process rank zero of a parallel job.)

It also says that it will return -1 when called by any process (rank?) not equal to 0.

libyogrt/doc/man/yogrt_remaining.3

Line 47 in 4346565

Returns -1 when called from any task other than task zero.

On lassen, it appears that yogrt_remaining() returns reasonable positive values when called from rank 1.

[tepperly]

I am not really asking you to remove functionality. We're just trying to write tests in LiDO based on the documented behavior. Perhaps, the best possible resolution would be a preprocessor macro (e.g., LIBYOGRT_WORKS_ON_ALL_RANKS) indicating whether yogrt_remaining() works on all ranks or only rank 0. This would allow us to adapt our tests to know whether to expect postive values on ranks not equal to zero or not.

[morrone]

lassen is using the lsf backend for remaining time. We could perhaps fix that backend to correctly return -1 on non-zero tasks. According to IBM documentation, there may be an environment variable LS_JOBPID that will tell us the process ID.

[morrone]

Presumably process means rank in an MPI context?

Technically, it is process rank 0 as determined by the job launcher. MPI bootstrap processes (almost?) always use this number to assign ranks in MPI_COMM_WORLD. libyogrt is not an MPI application, so it only knows what the launcher tells it, typically through environment variables. So it would entirely be possible for an alternate MPI communicator to not contain that process at all.