check_socket_compatibility fails with srun on more than one processor

[ltimmerman3]

Describe the bug Checking for socket compatibility currently requires running srun path/to/sparc/executable without the --name input which invokes stdout with/without -socket. This fails when the run command invokes more than one processor as only the 0th task exits, while all others hang.

To Reproduce

• ssh into slurm managed environment
• set ASE_SPARC_COMMAND to srun path/to/sparc
• run with socket mode = True (Fig 5 b run configuration) on a node with > 1 processors requested

Expected behavior All processes should exit

Actual output or error trace Only task 0 exits

This can be handled by enforcing srun -n 1 path/to/sparc as the run command for the compatibility check. Need to decide how to implement. Simplest: Check if "srun" in command -> edit command to be srun -n 1

[alchem0x2A]

that's indeed one issue with srun (as opposed to mpirun) that terminating srun processes require using slurm to terminate the step. https://github.com/SPARC-X/SPARC-X-API/blob/c419ce21b87fd8ddc795a5c7034be7a90162a641/sparc/calculator.py#L797 has implemented the termination procedure but there could be more things happening on the actual srun hierachy, I'll take a look

we could also implement closing the socket on receiving the EXIT message the C-SPARC side. This may be actually safer to work on, since enumerating all possible combinations of mpi/slurm is tedious.

To test

• [ ] Check if EXIT message is implemented in the C-SPARC socket
• [ ] Test on mpirun & srun on multiple processors
• [ ] Test on srun on multiple nodes