Geocoder Cache

[hannahblue]

Context

Currently we use the API for every geocode even if it's an address we've geocoded before. This is costly for both us and our customers. It is particularly problematic in sync tables where the entire table gets geocoded on every sync (issue here).

@antoniocarlon has begun developing the functionality to cache geocoding results so that new geocodes can query the cache first and then hit the data service provider for anything that wasn't in the cache (link).

To do

• [ ] Determine best table format
• [ ] Determine the best normalization mechanism for ids
• [ ] Refactor Python code (API key as a parameter for example)
• [ ] Refactor other (non Mapbox) geocoders to have the expected functions (may not be needed)
• [ ] Refactor server (PLPython) functions to use the proxy/cache

cc @antoniocarlon, @juanignaciosl

[juanignaciosl]

For address normalization, we could consider pypostal (which is taken into account by @andy-esch for CartoDB/cartoframes/issues/377).

[hannahblue]

Geocoding from cached geocodes is apparently against our Mapbox contract. Let's hold off on it for now.