pio configuration checks

[anton-seaice]

We should add checks to the pio initialisation modules to confirm that these options are valid for each model component and the mediator:

     pio_async_interface = .false.
     pio_netcdf_format = nothing
     pio_numiotasks = 5
     pio_rearranger = 1
     pio_root = 1
     pio_stride = 48
     pio_typename = netcdf4p

I think the checks would go here:
https://github.com/ESCOMP/CMEPS/blob/ffb573791c7670ada8660dbac96c6c84886f62d3/cesm/nuopc_cap_share/driver_pio_mod.F90#L366

But it may be they fit at a higher level within the CMEPS code.

The most likely error we would make it to request io from a PE which doesn't exist. i.e. we need to check

pio_numiotasks*pio_stride+pio_root<_ntasks

This would avoid isses like #156

Another error to check is if pio_typename is compatible with the pio version built.

[minghangli-uni]

pio_numiotasks*pio_stride+pio_root<_ntasks

For sequential runs, the above formula will break using the following setup, since the formula pio_numiotasks * pio_stride in this case already represents the total number of processors.

     pio_async_interface = .false.
     pio_netcdf_format = nothing
     pio_numiotasks = 5
     pio_rearranger = 1
     pio_root = 1
     pio_stride = 48
     pio_typename = netcdf4p

[anton-seaice]

For sequential runs, the above formula will break using the following setup, since the formula pio_numiotasks * pio_stride in this case already represents the total number of processors.

Yes -it should be (pio_numiotasks-1)* ...

It looks like there is some logic which just changes these values if they are wrong:

          if(.not. pio_comp_settings(i)%pio_async_interface) then
             call NUOPC_CompAttributeGet(gcomp(i), name="pio_stride", value=cval, rc=rc)
             if (chkerr(rc,__LINE__,u_FILE_u)) return
             read(cval, *) pio_comp_settings(i)%pio_stride
             if(pio_comp_settings(i)%pio_stride <= 0 .or. pio_comp_settings(i)%pio_stride > npets) then
                pio_comp_settings(i)%pio_stride = min(npets, default_stride)
             endif
          
             call NUOPC_CompAttributeGet(gcomp(i), name="pio_numiotasks", value=cval, rc=rc)
             if (chkerr(rc,__LINE__,u_FILE_u)) return
             read(cval, *) pio_comp_settings(i)%pio_numiotasks
          
             if(pio_comp_settings(i)%pio_numiotasks < 0 .or. pio_comp_settings(i)%pio_numiotasks > npets) then
                pio_comp_settings(i)%pio_numiotasks = max(1,npets/pio_comp_settings(i)%pio_stride)
             endif


             call NUOPC_CompAttributeGet(gcomp(i), name="pio_root", value=cval, rc=rc)
             if (chkerr(rc,__LINE__,u_FILE_u)) return
             read(cval, *) pio_comp_settings(i)%pio_root
          
             if(pio_comp_settings(i)%pio_root < 0 .or. pio_comp_settings(i)%pio_root > npets) then
                pio_comp_settings(i)%pio_root = 0
             endif
          endif

I guess we could add

if( (pio_comp_settings(i)%pio_root + (pio_comp_settings(i)%pio_numiotasks-1)*pio_comp_settings(i)%pio_stride > npets) then
                pio_comp_settings(i)%pio_stride = (npets-pio_comp_settings(i)%pio_root)/(pio_comp_settings(i)%pio_numiotasks-1)
             endif

Or throw an error if the condition fails. Shall we do this in the CESM share, or payu, or both ?

[minghangli-uni]

I think there is a bug here. See this payu-org/payu#496 (comment)

[anton-seaice]

Closed through payu-org/payu#496