search

ai2cm / fv3config

Manipulate FV3GFS run directories
Apache License 2.0
1 stars 0 forks source link

Include information about the runfile and/or docker image in the configuration #56

Closed nbren12 closed 4 years ago

nbren12 commented 4 years ago

When using the python wrapper, the runfile is an intrinsic component of the configuration of the model. It can update the state of the model in arbitrary ways. Therefore, it is probably important to include it in the configuration dictionary in some way. Likewise, the docker image is important provenance information that should be stored in the run directory. In addition, the runfile is actually needed to run the model, so could conceivably be provisioned using fv3config.

  1. fv3config writes information about the runfile and image to a file in the run-directory. This is important for understanding the provenance.
  2. Custom runfiles could be added to any Docker image based on fv3gfs-python. In the case of prognostic runs with ML, a "runfile" is not really distinct from "the model", so it would make sense to bundle them in one image together. This is not possible right now, since fv3run defaults to calling python -m mpi4py -m fv3gfs.run.
nbren12 commented 4 years ago

@mcgibbon and I discussed adding an environmental variable pointing to the runfile to allow baking the "runfile" into the dockerfile.