Closed altheaden closed 1 year ago
I need to do another look over the changes in the documentation, but I haven't yet been able to build the docs yet. I'm marking this PR in progress until I finalize that, but the framework changes are finished.
I tested this using the PR suite against a baseline, and everything passed as expected.
I made some modifications to the PR description. Feel free to further edit if they're not clear.
My testing passed on Chrysalis. I used a test-merge of this branch with main
and using main
as a baseline I ran:
ocean/baroclinic_channel/10km/default
ocean/baroclinic_channel/10km/decomp
ocean/baroclinic_channel/10km/restart
ocean/baroclinic_channel/10km/threads
ocean/baroclinic_channel/10km/rpe
ocean/global_convergence/qu/cosine_bell_with_viz
ocean/global_convergence/icos/cosine_bell_with_viz
ocean/inertial_gravity_wave/convergence
ocean/manufactured_solution/convergence
ocean/single_column/960km/cvmix
ocean/single_column/960km/ideal_age
The manufactured_solution
test case fails in the expected way because the convergence rate is less than desired. The other tests passed and results are bit-for-bit identical to the baseline.
@xylar I tried to re-word things to make them a bit more understandable, let me know if that's more in line with what you were thinking.
Also - I just built the docs and they look as expected.
@altheaden, I've assigned this to you. Merge when you're ready.
I'm changing the labels since the things changed here are all in polaris/ocean
or docs
. The "framework" for our purposes is anything in the polaris
package and not in one of the component subpackages.
Before this merge, we would compute the cell count using either a simple heuristic, or information from a previous step (e.g. opening the mesh file from the
init
step). In task parallel, we need to constrain the resources for all steps before any steps have run. This means won't have information (e.g. the exact cell count) from any previous steps when we need to constrain the resources for the given step at runtime. Therefore, we want to just use the heuristic every time instead of trying to read in the cell count from a file that won't exist.Checklist
Testing
comment in the PR documents testing used to verify the changes