Revert some COMPASS changes intended to support conda MPI #545
Conversation
This env will make sure we use the nompi variant of netcdf4
This reverts commit 8217097. As long as the `netcdf4` package used is the `nompi` variant, we don't need to detect MPI from the conda environment or call python scripts with `mpirun`.
This reverts commit 190338d. We do not need to run `planar_hex` with `mpirun` in any circumstances as long as we use the `nompi` variant of `netcdf4`
|
I just need to deploy |
|
Okay, |
There was a problem hiding this comment.
This passed the nightly regression suite using compass 0.1.5 on grizzly. I also created the other rpe_test domains with no problems. Thanks @xylar!
|
I'll merge in after Redi is completely settled on E3SM - probably in a few days. |
|
Retested on IC. Passes nightly regression suite using both compass 0.1.4 and 0.1.5. Runs e3sm_coupling successfully with compass 0.1.4. The e3sm_coupling steps with Merging in, as that specific problem is not caused by this PR. |
|
Thanks for merging these 3, @mark-petersen. I'm working on a fix for #564 as you see. |
…an/develop Revert some COMPASS changes intended to support conda MPI MPAS-Dev#545 It turns out not to be necessary to run python scripts with `mpirun` as long as we don't use the MPI version of `netcdf4`. This is possible even if we want to use the MPI version of `esmf` (which we do because `ESMF_RegridWeightGen` is way too slow in serial for certain mesh combinations). This PR reverts: * 8217097, which modified `setup_testcase.py` to add `mpirun` to python * scripts in certain circumstances * 190338d, which made sure `planar_hex` (which wasn't obviously a python * script) also got run in this way. This merge also updates to `compass` env. 0.1.5, which makes sure we always get the `nompi` variant of `netcdf4`, so that the shenanigans with `mpirun` isn't needed.
It turns out not to be necessary to run python scripts with
mpirunas long as we don't use the MPI version ofnetcdf4. This is possible even if we want to use the MPI version ofesmf(which we do becauseESMF_RegridWeightGenis way too slow in serial for certain mesh combinations).This PR reverts:
setup_testcase.pyto addmpirunto python scripts in certain circumstancesplanar_hex(which wasn't obviously a python script) also got run in this way.This merge also updates to
compassenv. 0.1.5, which makes sure we always get thenompivariant ofnetcdf4, so that the shenanigans withmpirunisn't needed.