Disentangling Parcels

[willirath]

@delandmeterp In Paris, we briefly talked about disentangling the structure of Parcels into essentially a bunch of microservices.

In this Gist https://gist.github.com/willirath/6943b136ad85ad83078f201316f9170b I've played with Dask actors to have a spatial domain (like a fieldset) on one worker, the random number generator on another worker and the particles on the front end.

(Note that this is more of a thought experiment to better understand the structure of the problem.)

[delandmeterp]

Hi Willi,

Sorry for the very late reply, I was on holiday and came back last week. Thanks! It's nice to see we can really reduce the Particle tracker to minimal independent modules.

You already know that what I don't really like is the use of scipy.interpolate.interp2d(), which does not work for us as it should. But in that same spirit, when I was in scipy I've started to write a vectorial interpolator, using the library of fbriol to locate the particles. Combining those pieces, we could have a simple and very modulable particle tracker.

We should list what is needed for a minimal program, how to cut it in modules as small as possible, and how to interconnect those modules.